{ "generated_at": "2026-09-22", "datasets": { "included": { "label": "Included cases", "description": "Normalized AI incident records selected for the main public dataset.", "records": [ { "error_id": "AIEL-2024-001", "error_title": "Air Canada chatbot misstated bereavement-fare refund eligibility", "ai_system_name": "Air Canada support chatbot (technical nature not established in the decision)", "deployer": "Air Canada", "domain": "consumer travel", "error_type": "policy misstatement", "error_description": "The decision records that Air Canada's support chatbot told a passenger they could request a reduced bereavement fare retroactively within 90 days of ticket issuance. The linked Bereavement travel page said the policy did not apply after travel was completed. The tribunal found Air Canada responsible for information on its website and liable for negligent misrepresentation; the decision does not establish the chatbot's technical architecture.", "canonical_source_conflicted": "Air Canada bereavement-fare policy", "mitigation_gap": "The tribunal found that Air Canada did not take reasonable care to ensure its chatbot was accurate. The decision does not identify the model architecture, technical guardrails, or escalation controls.", "reliance_or_harm": "The tribunal accepted that the passenger reasonably relied on the chatbot information, purchased two flights, and suffered C$650.88 in damages.", "public_matter_type": "tribunal claim", "public_matter_name": "Moffatt v. Air Canada, 2024 BCCRT 149", "filing_status": "ordered", "jurisdiction": "British Columbia Civil Resolution Tribunal", "error_date": "2022-11", "public_record_link": "https://decisions.civilresolutionbc.ca/crt/crtd/en/525448/1/document.do", "secondary_source_links": "https://canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html; https://dentons.com/insights/alerts/2024/february/29/air-canada-liable-for-negligent-misrepresentation-by-chatbot-on-website; https://canliiconnects.org/en/commentaries/93547", "confidence_score": "high", "notes_on_resolution": "The tribunal ordered Air Canada to pay C$812.02: C$650.88 in damages, C$36.14 in pre-judgment interest, and C$125 in tribunal fees. It dismissed the remaining claims and allowed post-judgment interest as applicable.", "tags": "consumer-facing; chatbot; policy-misstatement; negligent-misrepresentation; tribunal; small-claims; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "yes", "neutral_citation": "2024 BCCRT 149", "case_sameAs": "https://canlii.org/en/bc/bccrt/doc/2024/2024bccrt149/2024bccrt149.html", "legal_graph": { "proceedings": [ { "id": "aiel-2024-001-proceeding", "heard_by": [ "british-columbia-civil-resolution-tribunal" ], "parties": [ "aiel-2024-001-deployer" ], "determination_ids": [ "aiel-2024-001-determination" ], "matter_type": "small claims tribunal proceeding", "procedural_stage": "final-decision-issued" } ], "determinations": [ { "id": "aiel-2024-001-determination", "issued_by": [ "british-columbia-civil-resolution-tribunal" ], "issued_date": "2024-02-14", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2023-002", "error_title": "ChatGPT fabricated federal cases in an airline personal-injury brief", "ai_system_name": "ChatGPT", "deployer": "Levidow, Levidow & Oberman PC attorneys", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Counsel filed a brief containing nonexistent cases, false quotations, and false citations generated through ChatGPT. The lawyer then asked ChatGPT whether the cases were real and relied on the chatbot's answer instead of checking legal databases.", "canonical_source_conflicted": "Federal reporters, legal research databases, and actual case law", "mitigation_gap": "No independent cite-checking against legal databases; AI self-verification was substituted for source verification; supervising attorney signed without adequate review.", "reliance_or_harm": "Underlying lawsuit was dismissed and counsel/firm were sanctioned.", "public_matter_type": "sanctions order", "public_matter_name": "Mata v. Avianca, Inc., No. 22-cv-1461", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2023-06-22", "error_date": "2023-03-01", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.54.0_8.pdf", "secondary_source_links": "https://nytimes.com/2023/05/27/nyregion/avianca-airline-lawsuit-chatgpt.html", "confidence_score": "high", "notes_on_resolution": "The June 22, 2023 Opinion and Order on Sanctions imposed a $5,000 sanction jointly and severally on Peter LoDuca, Steven Schwartz, and Levidow, Levidow & Oberman P.C., and ordered notices to affected judges and the client. The indexed filing_date is the sanctions opinion and order filing date; the action entered federal court on removal on February 22, 2022.", "tags": "document-generation; legal-hallucination; sanctions-order; professional-discipline; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-01", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ], "legal_graph": { "authorities": [ { "id": "u-s-district-court-southern-district-of-new-york", "name": "U.S. District Court, Southern District of New York", "territorial_scope": [ "us-ny" ], "same_as": [ "https://wikidata.org/entity/Q673281" ] } ], "proceedings": [ { "id": "aiel-2023-002-proceeding", "heard_by": [ "u-s-district-court-southern-district-of-new-york" ], "filed_date": "2022-02-22", "parties": [], "determination_ids": [ "aiel-2023-002-determination" ], "matter_type": "federal civil action with ancillary sanctions proceeding", "procedural_stage": "federal-civil-action-after-removal" } ], "determinations": [ { "id": "aiel-2023-002-determination", "issued_by": [ "u-s-district-court-southern-district-of-new-york" ], "issued_date": "2023-06-22", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2024-003", "error_title": "ChatGPT-generated appellate filing cited a nonexistent case", "ai_system_name": "ChatGPT", "deployer": "Counsel for appellant in Second Circuit appeal", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A reply brief cited a case that did not exist. The Second Circuit identified the citation as apparently generated by ChatGPT and referred counsel to the court's grievance panel.", "canonical_source_conflicted": "Second Circuit case law and cited legal authority", "mitigation_gap": "No cite-checking caught the nonexistent case before filing; counsel did not ensure AI-assisted legal research matched real authority.", "reliance_or_harm": "Counsel was referred for possible discipline and the court issued a published warning on AI verification duties.", "public_matter_type": "sanctions / discipline referral order", "public_matter_name": "Park v. Kim, No. 22-2057", "filing_status": "ordered", "jurisdiction": "U.S. Court of Appeals for the Second Circuit", "filing_date": "2024-01-30", "error_date": "2023", "public_record_link": "https://govinfo.gov/content/pkg/USCOURTS-ca2-22-02057/pdf/USCOURTS-ca2-22-02057-0.pdf", "secondary_source_links": "https://law.nyu.edu/sites/default/files/Park%20v.%20Kim,%20No.%2022-2057%20(2024).pdf", "confidence_score": "high", "notes_on_resolution": "Matter referred to grievance panel; opinion emphasizes duty to verify AI-generated legal citations.", "tags": "document-generation; legal-hallucination; professional-discipline; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-01", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2025-004", "error_title": "AI-generated cases in Morgan & Morgan opposition brief", "ai_system_name": "ChatGPT / generative AI legal drafting workflow", "deployer": "Morgan & Morgan attorneys", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorneys filed briefing containing nonexistent or unsupported case citations attributed to generative AI use. Reporting and case-analysis sources describe a sanctions response after the court identified fabricated authority.", "canonical_source_conflicted": "Actual case law and court record", "mitigation_gap": "AI-generated legal citations were not verified against source materials before filing.", "reliance_or_harm": "Lawyers were fined and the firm faced sanctions-related scrutiny.", "public_matter_type": "sanctions order", "public_matter_name": "Wadsworth v. Walmart Inc.", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Wyoming", "filing_date": "2025", "error_date": "2025", "public_record_link": "https://app.ediscoveryassistant.com/case_law/62637-wadsworth-v-walmart-inc", "secondary_source_links": "https://news.bloomberglaw.com/litigation/morgan-morgan-lawyers-fined-for-hallucinated-ai-citations; https://abajournal.com/news/article/no-42-law-firm-by-headcount-could-face-sanctions-over-fake-case-citations-generated-by-chatgpt", "confidence_score": "medium", "notes_on_resolution": "Included as a notable sanctions event, but the primary court order should be downloaded and archived before treating as fully verified.", "tags": "document-generation; legal-hallucination; sanctions-order; needs-primary-record", "source_quality": "reliable reporting", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "yes", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-005", "error_title": "AI-drafted FCRA appellate reply brief fabricated quotations", "ai_system_name": "Casetext/CoCounsel and vLex (named; court-found; court did not conclusively pin which product generated hallucinations)", "deployer": "Heather Hersh / FCRA Attorneys", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A reply brief contained fabricated quotations and additional misrepresentations. The Fifth Circuit found that AI-generated material had been filed without adequate verification and that counsel's response to the show-cause order compounded the problem.", "canonical_source_conflicted": "Fifth Circuit record, quoted opinions, served Rule 11 materials", "mitigation_gap": "No independent verification of quotations or propositions before filing; show-cause response was evasive about AI authorship.", "reliance_or_harm": "Attorney sanctioned and ordered to pay $2,500 to the clerk.", "public_matter_type": "sanctions order", "public_matter_name": "Fletcher v. Experian Information Solutions, Inc., No. 25-20086", "filing_status": "sanctioned", "jurisdiction": "U.S. Court of Appeals for the Fifth Circuit", "filing_date": "2026-02-18", "error_date": "2025", "public_record_link": "https://www.ca5.uscourts.gov/opinions/pub/25/25-20086-CV0.pdf", "secondary_source_links": "https://reuters.com/legal/government/us-appeals-court-orders-lawyer-pay-2500-over-ai-hallucinations-brief-2026-02-18/", "confidence_score": "high", "notes_on_resolution": "Fifth Circuit Feb 18, 2026 final order: $2,500 sanction on appellate counsel Heather Hersh under inherent power; district-court counsel Shawn Jaffer Rule 11/$1927 sanctions vacated by separate order. Both Casetext/CoCounsel and vLex named as candidate products; Hersh admitted AI use when asked directly.", "tags": "document-generation; legal-hallucination; sanctions-order; professional-discipline; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-06-02", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2025-006", "error_title": "AI-generated contract-appeal brief invented cases and record citations", "ai_system_name": "Unspecified generative AI tool", "deployer": "Counsel for Huffman Construction, LLC", "domain": "legal services / government contracts", "error_type": "hallucinated authority", "error_description": "A post-hearing brief contained fictitious cases, false transcript citations, false Rule 4 file citations, and references to transcript pages that did not exist. Counsel admitted AI use in connection with the brief.", "canonical_source_conflicted": "ASBCA precedents, transcript volumes, and Rule 4 appeal file", "mitigation_gap": "Counsel and staff reviewed the brief but did not verify citations against the actual record or authorities; safeguards were found inadequate.", "reliance_or_harm": "Board struck the entire reply brief and denied leave to refile.", "public_matter_type": "sanctions / strike order", "public_matter_name": "Huffman Construction, LLC, ASBCA Nos. 62591, 62783", "filing_status": "ordered", "jurisdiction": "Armed Services Board of Contract Appeals", "filing_date": "2025-10-23", "error_date": "2025-08-15", "public_record_link": "https://www.asbca.mil/LinkClick.aspx?fileticket=eAIy2KSL_Zg%3D&portalid=143", "secondary_source_links": "https://smithlaw.com/newsroom/publications/you-cant-spell-sanction-without-a-and-i-when-unchecked-ai-hallucinations-result-in-court-sanctions; https://burr.com/government-contracting/gen-ai-misuse-in-procurement-litigation", "confidence_score": "high", "notes_on_resolution": "Reply brief struck; corrected reply denied; board issued broader AI-use caution.", "tags": "document-generation; legal-hallucination; sanctions-order; tribunal; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2025-007", "error_title": "AI-hallucinated class-settlement brief made counsel inadequate", "ai_system_name": "Six unspecified AI tools used concurrently", "deployer": "Plaintiff class counsel in Vuori wage-and-hour case", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Class counsel used multiple AI tools to draft and cross-check a class settlement approval brief. The filing contained fabricated quotations from real cases and at least one nonexistent case, and a corrected filing failed to remove the hallucinations.", "canonical_source_conflicted": "Actual case law and settlement approval standards", "mitigation_gap": "AI tools were used to verify each other instead of checking legal databases; corrected filing did not correct false authority.", "reliance_or_harm": "Settlement motions were struck; counsel was found inadequate as class counsel; class settlement collapsed into individual resolution.", "public_matter_type": "sanctions order", "public_matter_name": "Buchanan v. Vuori, Inc., No. 5:23-cv-01121-NC", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of California", "filing_date": "2025-11-20", "error_date": "2025-10-28", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.cand.409525/gov.uscourts.cand.409525.96.0.pdf", "secondary_source_links": "https://blogs.duanemorris.com/classactiondefense/2026/02/03/ai-hallucinated-case-citations-prompt-sanctions-and-delay-class-action-settlement/; https://vibegraveyard.ai/story/ndcal-dal-bon-ai-citation-class-action-sanctions/", "confidence_score": "high", "notes_on_resolution": "Court struck settlement motions with prejudice, imposed a monetary fine, and referred counsel for professional conduct review.", "tags": "document-generation; legal-hallucination; sanctions-order; professional-discipline; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-25", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2024-008", "error_title": "AI-fabricated citations contributed to attorney discipline", "ai_system_name": "Unspecified generative AI", "deployer": "Thomas Grant Neusom", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney filings contained fabricated AI-generated citations in a federal matter. The AI issue was part of a broader disciplinary record that included evasive responses and other noncompliance.", "canonical_source_conflicted": "Actual case law and federal court filing obligations", "mitigation_gap": "No citation verification; failure to correct when challenged; broader noncompliance with court and bar processes.", "reliance_or_harm": "Attorney was suspended in federal court and later disciplined by the Florida Supreme Court.", "public_matter_type": "disciplinary matter", "public_matter_name": "In re Thomas Grant Neusom / The Florida Bar v. Neusom", "filing_status": "resolved", "jurisdiction": "U.S. District Court, Middle District of Florida; Florida Supreme Court", "filing_date": "2024-03-08", "error_date": "2023", "public_record_link": "https://ecf.flmd.uscourts.gov/cgi-bin/show_public_doc?2024-00002-6-2-mc", "secondary_source_links": "https://lawnext.com/wp-content/uploads/2024/03/M.D.-Fla.-24-mc-00002-dckt-000003_000-filed-2024-02-01.pdf; https://jcorsmeier.wordpress.com/2024/03/20/florida-middle-district-federal-judge-suspends-florida-lawyer-for-filing-false-cases-created-by-artificial-intelligence/", "confidence_score": "high", "notes_on_resolution": "AI hallucinations are recorded as corroborating misconduct, not necessarily the sole basis for discipline.", "tags": "document-generation; legal-hallucination; professional-discipline; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ], "legal_graph": { "authorities": [ { "id": "u-s-district-court-middle-district-of-florida", "name": "U.S. District Court, Middle District of Florida", "territorial_scope": [ "us-fl" ], "same_as": [ "https://wikidata.org/entity/Q1384933" ] } ], "proceedings": [ { "id": "aiel-2024-008-proceeding", "heard_by": [ "u-s-district-court-middle-district-of-florida" ], "determination_ids": [ "aiel-2024-008-determination" ], "procedural_stage": "federal-discipline" } ], "determinations": [ { "id": "aiel-2024-008-determination", "issued_by": [ "u-s-district-court-middle-district-of-florida" ], "issued_date": "2024-03-08", "disposition": "confirmed" } ], "retired_identifiers": [ { "kind": "authority", "id": "u-s-district-court-middle-district-of-florida-florida-supreme-court", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated separate federal and state courts." } ] } }, { "error_id": "AIEL-2023-009", "error_title": "Automated tutor-application screen rejected older applicants", "ai_system_name": "iTutorGroup application screening software", "deployer": "iTutorGroup, Inc. and affiliates", "domain": "employment", "error_type": "discriminatory output", "error_description": "Application software automatically rejected female applicants age 55 or older and male applicants age 60 or older. The EEOC alleged the screening logic unlawfully denied older applicants an employment opportunity.", "canonical_source_conflicted": "Age Discrimination in Employment Act", "mitigation_gap": "No pre-deployment age-discrimination audit or human override prevented hard-coded age thresholds from eliminating qualified applicants.", "reliance_or_harm": "More than 200 applicants were allegedly rejected; company settled with EEOC for $365,000 and injunctive relief.", "public_matter_type": "regulatory enforcement / federal lawsuit", "public_matter_name": "EEOC v. iTutorGroup, Inc., et al., No. 1:22-cv-02565", "filing_status": "settled", "jurisdiction": "EEOC; U.S. District Court, Eastern District of New York", "filing_date": "2022-05-05", "error_date": "2020-03", "public_record_link": "https://eeoc.gov/newsroom/itutorgroup-pay-365000-settle-eeoc-discriminatory-hiring-suit", "secondary_source_links": "https://seyfarth.com/news-insights/eeocs-settlement-challenging-simple-algorithm-provides-warning-for-employers-using-artificial-intelligence.html", "confidence_score": "high", "notes_on_resolution": "EEOC announced $365,000 settlement and remedial measures.", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/bias-prevention.json" ], "tags": "internal-use-but-public-consequence; employment; discrimination-claim; agency-action; public-record-confirmed", "source_quality": "official statement", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "legal_graph": { "authorities": [ { "id": "u-s-district-court-eastern-district-of-new-york", "name": "U.S. District Court, Eastern District of New York", "territorial_scope": [ "us-ny" ] } ], "proceedings": [ { "id": "aiel-2023-009-proceeding", "heard_by": [ "u-s-district-court-eastern-district-of-new-york" ], "determination_ids": [ "aiel-2023-009-determination" ] } ], "determinations": [ { "id": "aiel-2023-009-determination", "issued_by": [ "u-s-district-court-eastern-district-of-new-york" ], "issued_date": "2023-09-11", "disposition": "settled" } ], "retired_identifiers": [ { "kind": "authority", "id": "eeoc-u-s-district-court-eastern-district-of-new-york", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated the enforcing agency with the court hearing the case." } ] } }, { "error_id": "AIEL-2024-010", "error_title": "SafeRent Score allegedly downgraded housing-voucher applicants", "ai_system_name": "SafeRent Score", "deployer": "SafeRent Solutions, LLC / tenant-screening users", "domain": "housing", "error_type": "discriminatory output", "error_description": "Tenant-screening score allegedly relied heavily on credit history and non-tenancy debt in a way that disadvantaged Black and Hispanic rental applicants using housing vouchers. DOJ and HUD filed a statement of interest supporting Fair Housing Act disparate-impact applicability to algorithmic tenant screening.", "canonical_source_conflicted": "Fair Housing Act and voucher-backed rent-payment facts", "mitigation_gap": "No adequate fair-housing disparate-impact testing; model allegedly omitted or undervalued voucher payment certainty.", "reliance_or_harm": "Voucher holders allegedly denied housing; settlement created monetary and injunctive relief.", "public_matter_type": "court complaint / agency statement of interest", "public_matter_name": "Louis et al. v. SafeRent Solutions, LLC et al.", "filing_status": "settled", "jurisdiction": "U.S. District Court, District of Massachusetts; DOJ; HUD", "filing_date": "2022-05", "error_date": "2022", "public_record_link": "https://justice.gov/crt/case/louis-et-al-v-saferent-et-al-d-mass", "secondary_source_links": "https://justice.gov/archives/opa/pr/justice-department-files-statement-interest-fair-housing-act-case-alleging-unlawful-algorithm; https://cohenmilstein.com/rental-applicants-reach-2-28m-settlement-agreement-for-discriminatory-ai-powered-screening-tool/", "confidence_score": "high", "notes_on_resolution": "Settlement announced in 2024; algorithmic tenant screening treated as subject to fair-housing law.", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/bias-prevention.json" ], "tags": "internal-use-but-public-consequence; housing; discrimination-claim; agency-action; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "legal_graph": { "authorities": [ { "id": "u-s-district-court-district-of-massachusetts", "name": "U.S. District Court, District of Massachusetts", "territorial_scope": [ "us-ma" ] } ], "proceedings": [ { "id": "aiel-2024-010-proceeding", "heard_by": [ "u-s-district-court-district-of-massachusetts" ], "determination_ids": [ "aiel-2024-010-determination" ] } ], "determinations": [ { "id": "aiel-2024-010-determination", "issued_by": [ "u-s-district-court-district-of-massachusetts" ], "disposition": "settled" } ], "retired_identifiers": [ { "kind": "authority", "id": "u-s-district-court-district-of-massachusetts-doj-hud", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated the court with participating federal agencies." } ] } }, { "error_id": "AIEL-2022-011", "error_title": "MiDAS automated fraud determinations falsely accused unemployment claimants", "ai_system_name": "Michigan Integrated Data Automated System (MiDAS)", "deployer": "Michigan Unemployment Insurance Agency", "domain": "government services", "error_type": "false fraud classification", "error_description": "MiDAS issued automated unemployment-fraud determinations that were later found overwhelmingly erroneous. Claimants were assessed penalties and collection actions after automated determinations without adequate process.", "canonical_source_conflicted": "Unemployment insurance fraud standards and constitutional due-process requirements", "mitigation_gap": "Automated flags lacked meaningful human review and adequate notice/opportunity to contest; error monitoring and escalation failed.", "reliance_or_harm": "Thousands of claimants were wrongfully accused and faced garnishments, penalties, and collection actions; state settled for $20 million.", "public_matter_type": "class action / state constitutional tort", "public_matter_name": "Bauserman v. Unemployment Insurance Agency", "filing_status": "settled", "jurisdiction": "Michigan Court of Claims; Michigan Supreme Court", "filing_date": "2015", "error_date": "2013-10", "public_record_link": "https://michigan.gov/ag/news/press-releases/2022/10/20/som-settlement-of-civil-rights-class-action-alleging-false-accusations-of-unemployment-fraud", "secondary_source_links": "https://law.justia.com/cases/michigan/supreme-court/2022/160813.html; https://btah.org/case-study/michigan-unemployment-insurance-false-fraud-determinations.html", "confidence_score": "high", "notes_on_resolution": "Michigan announced $20 million settlement; Michigan Supreme Court recognized damages availability for the constitutional tort theory.", "tags": "internal-use-but-public-consequence; government-benefits; eligibility-error; misclassification; public-record-confirmed", "source_quality": "official statement", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json", "https://everyailaw.com/obligation-category/explainability.json", "https://everyailaw.com/obligation-category/bias-prevention.json" ], "legal_graph": { "authorities": [ { "id": "michigan-court-of-claims", "name": "Michigan Court of Claims", "territorial_scope": [ "us-mi" ] }, { "id": "michigan-supreme-court", "name": "Michigan Supreme Court", "territorial_scope": [ "us-mi" ] } ], "proceedings": [ { "id": "aiel-2022-011-court-of-claims-proceeding", "title": "Bauserman v. Unemployment Insurance Agency, Court of Claims No. 15-202-MM", "filed_date": "2015-09-09", "heard_by": [ "michigan-court-of-claims" ], "determination_ids": [], "procedural_stage": "trial" }, { "id": "aiel-2022-011-proceeding", "heard_by": [ "michigan-supreme-court" ], "determination_ids": [ "aiel-2022-011-determination" ], "procedural_stage": "appeal" } ], "determinations": [ { "id": "aiel-2022-011-determination", "issued_by": [ "michigan-supreme-court" ], "issued_date": "2022-07-26", "disposition": "partial" } ], "retired_identifiers": [ { "kind": "authority", "id": "michigan-court-of-claims-michigan-supreme-court", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated trial and appellate courts." } ] } }, { "error_id": "AIEL-2017-012", "error_title": "RUGs care-allocation algorithm cut disabled beneficiaries' Medicaid hours", "ai_system_name": "Resource Utilization Groups (RUGs) computerized care-allocation algorithm", "deployer": "Arkansas Department of Human Services", "domain": "government services / healthcare", "error_type": "wrongful eligibility / allocation", "error_description": "Arkansas replaced nurse assessment logic with a computerized RUGs allocation model that reduced home-care hours for many Medicaid beneficiaries. Litigation alleged the system failed to account for individualized conditions and was implemented without proper rulemaking or due process.", "canonical_source_conflicted": "Medicaid home- and community-based services rules, state APA procedures, individualized care needs", "mitigation_gap": "No adequate rulemaking, transparency, individualized validation, or notice process accompanied the algorithmic allocation change.", "reliance_or_harm": "Disabled beneficiaries lost care hours and challenged the system; state-court rulings and later federal litigation led to relief and reforms.", "public_matter_type": "state APA / civil rights litigation", "public_matter_name": "Ledgerwood v. Arkansas Department of Human Services / related federal litigation", "filing_status": "resolved", "jurisdiction": "Arkansas Supreme Court; U.S. District Court, Eastern District of Arkansas; Eighth Circuit", "filing_date": "2017-01-27", "error_date": "2016", "public_record_link": "https://arcourts.gov/sites/default/files/Appellate%20Update%20November%202017.pdf", "secondary_source_links": "https://humanservices.arkansas.gov/divisions-shared-services/aging-adult-behavioral-health-services/find-home-community-based-services-for-adults-seniors/archoices-in-homecare/court-ordered-information/; https://btah.org/case-study/arkansas-medicaid-home-and-community-based-services-hours-cuts.html", "confidence_score": "high", "notes_on_resolution": "Algorithm invalidated on rulemaking grounds; later proceedings addressed successor systems and systemic relief.", "tags": "internal-use-but-public-consequence; government-benefits; healthcare; eligibility-error; due-process; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-01", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json", "https://everyailaw.com/obligation-category/explainability.json" ], "legal_graph": { "authorities": [ { "id": "arkansas-supreme-court", "name": "Arkansas Supreme Court", "territorial_scope": [ "us-ar" ] } ], "proceedings": [ { "id": "aiel-2017-012-proceeding", "heard_by": [ "arkansas-supreme-court" ], "determination_ids": [ "aiel-2017-012-determination" ], "procedural_stage": "state-appeal" } ], "determinations": [ { "id": "aiel-2017-012-determination", "issued_by": [ "arkansas-supreme-court" ], "disposition": "confirmed" } ], "retired_identifiers": [ { "kind": "authority", "id": "arkansas-supreme-court-u-s-district-court-eastern-district-of-arkansas-eighth-circuit", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated state and federal procedural stages." } ] } }, { "error_id": "AIEL-2023-013", "error_title": "Rite Aid facial-recognition system generated false shoplifting matches", "ai_system_name": "Rite Aid facial-recognition surveillance system", "deployer": "Rite Aid", "domain": "retail / consumer protection", "error_type": "false match", "error_description": "FTC alleged Rite Aid deployed facial recognition in stores and generated false-positive matches that disproportionately affected certain consumers, including people of color. Employees allegedly acted on false alerts by surveilling, searching, removing, or reporting customers.", "canonical_source_conflicted": "Accurate identity evidence and consumer-protection obligations", "mitigation_gap": "Insufficient testing, monitoring, bias controls, employee training, and consumer notice before acting on face-recognition alerts.", "reliance_or_harm": "Consumers were allegedly wrongly accused, embarrassed, or subjected to law-enforcement contact; FTC order banned Rite Aid from using facial recognition for five years.", "public_matter_type": "regulatory enforcement", "public_matter_name": "FTC v. Rite Aid facial-recognition enforcement", "filing_status": "settled", "jurisdiction": "U.S. Federal Trade Commission", "filing_date": "2023-12-19", "error_date": "2012-2020", "public_record_link": "https://ftc.gov/news-events/news/press-releases/2023/12/rite-aid-banned-using-ai-facial-recognition-after-ftc-says-retailer-deployed-technology-without", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "FTC announced five-year facial-recognition ban and algorithmic fairness/monitoring requirements.", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/bias-prevention.json" ], "tags": "facial-recognition; false-match; consumer-protection; agency-action; public-record-confirmed", "source_quality": "official statement", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no" }, { "error_id": "AIEL-2024-014", "error_title": "Facial-recognition false match led to Robert Williams arrest", "ai_system_name": "Michigan State Police / DataWorks Plus facial-recognition pipeline", "deployer": "Detroit Police Department and Michigan State Police", "domain": "criminal justice", "error_type": "false match", "error_description": "A grainy surveillance still was run through facial recognition and returned Robert Williams as a candidate. Police allegedly relied on the output and a flawed photo identification process, leading to a wrongful arrest for a theft he did not commit.", "canonical_source_conflicted": "Actual suspect identity and warrant probable-cause standards", "mitigation_gap": "No policy requiring independent corroboration before arrest; warrant materials did not adequately disclose the limits of the face-recognition lead.", "reliance_or_harm": "Williams was arrested, detained, and later brought a civil-rights suit; settlement included damages and policy reforms.", "public_matter_type": "federal civil rights lawsuit", "public_matter_name": "Williams v. City of Detroit, No. 2:21-cv-10827", "filing_status": "settled", "jurisdiction": "U.S. District Court, Eastern District of Michigan", "filing_date": "2021-04", "error_date": "2020-01", "public_record_link": "https://aclu.org/cases/williams-v-city-of-detroit-face-recognition-false-arrest", "secondary_source_links": "https://freep.com/story/news/local/michigan/detroit/2024/06/28/man-wrongfully-arrested-with-facial-recognition-tech-settles-lawsuit/74243839007/; https://aclumich.org/cases/facial-recognition/", "confidence_score": "high", "notes_on_resolution": "Settlement included $300,000 and Detroit Police Department facial-recognition reforms.", "tags": "facial-recognition; false-match; wrongful-arrest; civil-rights; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/bias-prevention.json", "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2024-015", "error_title": "Facial-recognition lead contributed to Nijeer Parks detention", "ai_system_name": "Out-of-state facial-recognition search run by a Rockland County (N.Y.) Sheriff's investigator via the New York State Intelligence Center; IDEMIA, Inc. pleaded as maker of the software (Clearview AI is not identified as the system used in the retrieved filings)", "deployer": "Woodbridge Police Department and intelligence-center intermediaries", "domain": "criminal justice", "error_type": "false match", "error_description": "A blurry fake-ID image was run through facial recognition and returned Nijeer Parks as a possible match. Police allegedly obtained a warrant without adequately verifying alibi evidence or disclosing the uncertainty of the technology.", "canonical_source_conflicted": "Actual suspect identity, alibi evidence, and warrant probable-cause standards", "mitigation_gap": "No adequate alibi check, DNA/fingerprint corroboration, or transparent disclosure of FRT provenance before arrest.", "reliance_or_harm": "Parks spent about 10 days in jail before charges were dismissed; civil-rights litigation followed, and the Township of Woodbridge settled for $300,000 in 2024 without admitting liability.", "public_matter_type": "federal civil rights lawsuit", "public_matter_name": "Parks v. McCormac, No. 2:21-cv-04021 (D.N.J.)", "filing_status": "settled", "jurisdiction": "U.S. District Court, District of New Jersey", "filing_date": "2022-09", "error_date": "2019-02", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.njd.462874/gov.uscourts.njd.462874.125.0.pdf", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.njd.462874/gov.uscourts.njd.462874.126.0.pdf; https://assets.aclu.org/live/uploads/2024/01/113-1.-ACLU-ACLU-NJ-Amicus-Brief-Filed.pdf; https://aclu.org/cases/parks-v-mccormac; https://aclu-nj.org/press-releases/aclu-nj-and-aclu-national-file-amicus-challenge-wrongful-arrest-due-face-recognition/; https://nytimes.com/2020/12/29/technology/facial-recognition-misidentify-jail.html", "confidence_score": "high", "notes_on_resolution": "At a 2024-07-08 settlement conference the parties reached a settlement in principle, and Magistrate Judge Wettre administratively terminated the action on 2024-07-09 by a 60-day order that is expressly not a Rule 41 dismissal (Dkt. 125). On 2024-09-05 defense counsel filed the executed Settlement Agreement and Release, under which the Township of Woodbridge pays $300,000 with no admission of liability, together with stipulations of dismissal (Dkt. 126). No merits ruling on the facial-recognition claims was issued; defendants' summary-judgment motion was pending when the case settled. An entered dismissal order after Dkt. 126 was not located.", "tags": "facial-recognition; false-match; wrongful-arrest; civil-rights; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "yes", "legal_graph": { "authorities": [ { "id": "u-s-district-court-district-of-new-jersey", "name": "U.S. District Court, District of New Jersey", "territorial_scope": [ "us-nj" ], "same_as": [ "https://wikidata.org/entity/Q775899" ] } ], "proceedings": [ { "id": "aiel-2024-015-proceeding", "title": "Parks v. McCormac, No. 2:21-cv-04021", "heard_by": [ "u-s-district-court-district-of-new-jersey" ], "determination_ids": [ "aiel-2024-015-determination" ], "matter_type": "federal civil rights action", "procedural_stage": "federal-civil-action-after-removal" } ], "determinations": [ { "id": "aiel-2024-015-determination", "issued_by": [ "u-s-district-court-district-of-new-jersey" ], "issued_date": "2024-07-09", "disposition": "settled" } ] } }, { "error_id": "AIEL-2025-016", "error_title": "Facial-recognition-seeded lineup led to Porcha Woodruff arrest", "ai_system_name": "Detroit Police Department facial-recognition system", "deployer": "Detroit Police Department", "domain": "criminal justice", "error_type": "false match", "error_description": "Detroit police used a facial-recognition lead to place Porcha Woodruff in a lineup for a carjacking investigation. The suit alleges obvious exculpatory circumstances, including pregnancy, were not adequately reconciled before arrest.", "canonical_source_conflicted": "Actual suspect identity and warrant probable-cause standards", "mitigation_gap": "FRT output seeded the lineup without sufficient independent corroboration or review of distinguishing characteristics.", "reliance_or_harm": "Woodruff was arrested while pregnant, detained, and later sued after charges were dropped.", "public_matter_type": "federal civil rights lawsuit", "public_matter_name": "Woodruff v. Oliver, No. 2:23-cv-11597", "filing_status": "dismissed / appeal flagged", "jurisdiction": "U.S. District Court, Eastern District of Michigan", "filing_date": "2023-07", "error_date": "2023-02", "public_record_link": "https://aclu.org/cases/woodruff-v-oliver", "secondary_source_links": "https://democracynow.org/2023/8/9/porcha_woodruff_false_facial_recognition_arrest; https://cbsnews.com/detroit/news/woman-wrongly-accused-carjacking-loses-lawsuit-detroit-police-used-facial-tech/", "confidence_score": "high", "notes_on_resolution": "Case reported dismissed at summary judgment in 2025 with appeal planned; keep procedural status under watch.", "tags": "facial-recognition; false-match; wrongful-arrest; civil-rights; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/bias-prevention.json", "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2024-017", "error_title": "Retail facial-recognition match allegedly led to Harvey Murphy's arrest", "ai_system_name": "Macy's / EssilorLuxottica facial-recognition workflow", "deployer": "EssilorLuxottica / Macy's loss-prevention personnel and Houston police", "domain": "retail / criminal justice", "error_type": "false match", "error_description": "A private retail loss-prevention workflow allegedly matched robbery footage to old mugshots of Harvey Murphy and communicated certainty to police. Murphy alleged the match and lineup process caused his arrest despite not being the robber.", "canonical_source_conflicted": "Actual suspect identity and reliable biometric-comparison standards", "mitigation_gap": "Private FRT comparison lacked qualified-examiner controls; old photographs and employee pre-briefing allegedly contaminated the identification process; police allegedly did not independently investigate.", "reliance_or_harm": "Murphy was detained and alleges severe jailhouse harm; civil case seeks damages.", "public_matter_type": "state civil lawsuit", "public_matter_name": "Murphy v. EssilorLuxottica USA Inc. et al., No. 2024-03265", "filing_status": "pending", "jurisdiction": "Harris County, Texas state court; remanded from S.D. Tex.", "filing_date": "2024-01-18", "error_date": "2022-10", "public_record_link": "https://govinfo.gov/content/pkg/USCOURTS-txsd-4_24-cv-00801/pdf/USCOURTS-txsd-4_24-cv-00801-1.pdf", "secondary_source_links": "https://govinfo.gov/content/pkg/USCOURTS-txsd-4_24-cv-00801/pdf/USCOURTS-txsd-4_24-cv-00801-0.pdf; https://regmedia.co.uk/2024/01/23/harvey_eugene_murphy_jr_lawsuit.pdf; https://dockets.justia.com/docket/texas/txsdce/4:2024cv00801/1952110; https://cbsnews.com/news/facial-recognition-mistaken-identity-sunglass-hut-robber-harvey-eugene-murphy-suing-after-sexually-assaulted-jail/", "confidence_score": "high", "notes_on_resolution": "Filed 2024-01-18 in the 125th Judicial District Court of Harris County, Texas (Cause No. 2024-03265) and removed to the Southern District of Texas as No. 4:24-cv-00801 on 2024-03-04. Magistrate Judge Edison's memorandum and recommendation of 2024-07-18 (Dkt. 26) recounts the petition's allegation that loss-prevention staff convinced police that artificial intelligence and facial recognition software identified Murphy, and recommends remand; Judge Hanks adopted it and remanded the case to the state court on 2024-08-14 (Dkt. 30). Murphy abandoned his claims against Stites and his negligence claims against Rodriguez in the remand motion. The federal orders decide jurisdiction only and make no finding on the facial-recognition allegations. The post-remand Harris County docket has not been observed.", "tags": "facial-recognition; false-match; wrongful-arrest; consumer-protection; civil-rights; public-record-confirmed", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "yes", "legal_graph": { "authorities": [ { "id": "harris-county-125th-judicial-district-court", "name": "Harris County 125th Judicial District Court", "territorial_scope": [ "us-tx" ] }, { "id": "u-s-district-court-southern-district-of-texas", "name": "U.S. District Court, Southern District of Texas", "territorial_scope": [ "us-tx" ] } ], "proceedings": [ { "id": "aiel-2024-017-proceeding", "heard_by": [ "harris-county-125th-judicial-district-court" ], "determination_ids": [], "procedural_stage": "state-trial" }, { "id": "aiel-2024-017-federal-removal-proceeding", "title": "Murphy v. EssilorLuxottica USA Inc. et al., No. 4:24-cv-00801", "filed_date": "2024-03-04", "heard_by": [ "u-s-district-court-southern-district-of-texas" ], "determination_ids": [], "procedural_stage": "federal-removal" } ], "determinations": [], "retired_identifiers": [ { "kind": "authority", "id": "harris-county-texas-state-court-remanded-from-s-d-tex", "former_type": "of:Authority", "reason": "Retired composite identifier that conflated the state proceeding with the federal removal stage." } ] } }, { "error_id": "AIEL-2025-018", "error_title": "Workday applicant-screening tools allegedly rejected older applicants at scale", "ai_system_name": "Workday HiredScore / Candidate Skills Match tools", "deployer": "Workday, Inc. as HR software vendor", "domain": "employment", "error_type": "discriminatory output", "error_description": "Plaintiff alleges Workday's AI screening tools caused discriminatory job-application rejections based on age, race, and disability-related factors. The case is included as a vendor-liability candidate because the public record identifies AI screening and a court allowed key claims to proceed, but discrete error rows may need to be split by plaintiff or application event.", "canonical_source_conflicted": "ADEA, Title VII, ADA, FEHA, and employer-specific applicant records", "mitigation_gap": "Alleged lack of disparate-impact audit, biased training/preferences, and absence of meaningful human review before rejection.", "reliance_or_harm": "Plaintiff alleges repeated job rejections through Workday systems; ADEA collective certification was reported.", "public_matter_type": "federal class / collective action", "public_matter_name": "Mobley v. Workday, Inc., No. 3:23-cv-00770", "filing_status": "pending", "jurisdiction": "U.S. District Court, Northern District of California", "filing_date": "2023-02-21", "error_date": "2018-2023", "public_record_link": "https://govinfo.gov/content/pkg/USCOURTS-cand-3_23-cv-00770/pdf/USCOURTS-cand-3_23-cv-00770-1.pdf", "secondary_source_links": "https://clearinghouse.net/case/44074/; https://fisherphillips.com/en/insights/insights/discrimination-lawsuit-over-workdays-ai-hiring-tools-can-proceed-as-class-action-6-things", "confidence_score": "medium", "notes_on_resolution": "Included with review flag because the public case is broader than a single bounded error event.", "tags": "internal-use-but-public-consequence; employment; discrimination-claim; decision-support; needs-event-splitting", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-04-22", "needs_review": "yes" }, { "error_id": "AIEL-2026-019", "error_title": "IBM algorithmic rehire screen allegedly blocked applicant due to age", "ai_system_name": "Unnamed IBM algorithmic hiring/rehire screening tool", "deployer": "International Business Machines Corporation", "domain": "employment", "error_type": "discriminatory output", "error_description": "Plaintiff, a 24-year IBM veteran, alleges an algorithmic hiring/rehire screening tool blocked his rehire at age 48, constituting age discrimination. Filed as Swanson v. International Business Machines Corporation in the Western District of Texas on 2026-05-23. The precise technical nature of the tool (machine-learning vs. rules-based) is not yet established on the public record.", "canonical_source_conflicted": "ADEA and applicant-specific rehire eligibility records", "mitigation_gap": "Alleged absence of meaningful human review of an automated rehire-eligibility screen and no disparate-impact safeguard against age-based exclusion.", "reliance_or_harm": "Plaintiff alleges denial of rehire after 24 years of prior IBM employment, attributed to an automated screen.", "public_matter_type": "federal lawsuit", "public_matter_name": "Swanson v. International Business Machines Corporation, No. 1:26-cv-01382", "filing_status": "filed", "jurisdiction": "U.S. District Court, Western District of Texas", "filing_date": "2026-05-23", "error_date": "2026", "public_record_link": "https://courtlistener.com/docket/73387361/swanson-v-international-business-machines-corporation/", "secondary_source_links": "https://hcamag.com/us/news/general/24-year-ibm-veteran-sues-says-algorithm-blocked-his-rehire-at-48/576441", "confidence_score": "medium", "notes_on_resolution": "Included by steward at early/filed posture. NOTE: AI system unnamed and ML-vs-rules nature unverified (admission gate 1 partial); complaint not yet pulled from PACER. Verify the tool is genuinely AI and re-confirm on next refresh. Sibling matter: John Loeffler v. IBM, No. 2:25-cv-05765 (C.D. Cal.), same defendant + age/algorithm-hiring theory.", "tags": "internal-use-but-public-consequence; employment; age-discrimination; hiring-screen; algorithmic-denial; needs-complaint-verification", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-06-01", "needs_review": "yes", "legal_graph": { "determinations": [], "retired_identifiers": [ { "kind": "determination", "id": "aiel-2026-019-determination", "former_type": "of:Determination", "reason": "Retired because a filed complaint is an unresolved allegation, not an adjudicative Determination." } ] } }, { "error_id": "AIEL-2026-020", "error_title": "CNN sues Perplexity over alleged use of its content in AI answers", "ai_system_name": "Perplexity answer engine + GenAI products (PerplexityBot / Perplexity-User crawlers; \"AI-First\" search index; RAG)", "deployer": "Perplexity AI, Inc.", "domain": "media / content", "error_type": "unauthorized content use; false attribution", "error_description": "CNN (a Warner Bros. Discovery subsidiary) sued Perplexity AI in the Southern District of New York (No. 1:26-cv-04427, filed 2026-05-28; jury demanded). The complaint alleges a two-stage harm: at the input stage, Perplexity’s \"PerplexityBot\" and \"Perplexity-User\" crawlers copy CNN content (over 17,000 works) to build an \"AI-First\" search index; at the output stage, Perplexity’s GenAI products (answer engine, enterprise chatbot, APIs, Comet browser) generate verbatim or substantially similar reproductions. CNN brings Copyright Act (17 U.S.C. § 101 et seq.) and Lanham Act claims, the latter including \"confabulated or ‘hallucinated’ content that Perplexity falsely attributes to CNN.\"", "canonical_source_conflicted": "CNN copyrighted works (17,000+ registrations) and CNN trademarks; 17 U.S.C. § 501 and the Lanham Act", "mitigation_gap": "Alleged crawling/copying of copyrighted content without licensing, and AI output that fabricates and falsely attributes content to CNN without adequate provenance or attribution controls.", "reliance_or_harm": "CNN alleges licensing/market harm from large-scale reproduction plus consumer confusion and brand dilution from hallucinated content falsely attributed to CNN.", "public_matter_type": "federal lawsuit", "public_matter_name": "Cable News Network, Inc. v. Perplexity AI, Inc., No. 1:26-cv-04427", "filing_status": "filed", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2026-05-28", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.664916/gov.uscourts.nysd.664916.1.0_1.pdf", "secondary_source_links": "https://courtlistener.com/docket/73402641/cable-news-network-inc-v-perplexity-ai-inc/", "confidence_score": "high", "notes_on_resolution": "Complaint (Doc 1, 54 pp) pulled from RECAP and verified 2026-06-01. Copyright + Lanham (trademark/false-attribution) claims; part of a multi-publisher wave against Perplexity (NYT, Chicago Tribune, Encyclopaedia Britannica/Merriam-Webster, Reddit, Dow Jones).", "tags": "generative-ai; copyright; lanham-act; trademark; content-reproduction; hallucinated-attribution; rag; media", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-06-01", "needs_review": "yes", "legal_graph": { "determinations": [], "retired_identifiers": [ { "kind": "determination", "id": "aiel-2026-020-determination", "former_type": "of:Determination", "reason": "Retired because a filed complaint is an unresolved allegation, not an adjudicative Determination." } ] } }, { "error_id": "AIEL-2026-021", "error_title": "Oklahoma attorney publicly reprimanded for ChatGPT-fabricated citations", "ai_system_name": "ChatGPT", "deployer": "Attorney respondent (Oklahoma bar disciplinary proceeding)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "In reciprocal attorney discipline, the Oklahoma Supreme Court publicly reprimanded attorney Matthew Brett Reeves (2026 OK 37, decided 2026-05-27) for filing five fabricated citations across two motions in the underlying federal matter Johnson v. Dunn, No. 2:21-cv-1701-AMM (N.D. Ala.). Reeves admitted the citations were \"hallucinated by ChatGPT\" and that he failed to verify them, violating his firm’s AI policy. The OBA noted this is the first reported Oklahoma attorney-discipline case involving AI-fabricated citations.", "canonical_source_conflicted": "Genuine case law and the duty of candor to the tribunal", "mitigation_gap": "No verification of AI-generated citations against a real reporter before filing.", "reliance_or_harm": "Fabricated authorities submitted to a court; attorney received a public reprimand.", "public_matter_type": "attorney disciplinary order", "public_matter_name": "State of Oklahoma ex rel. Oklahoma Bar Association v. Reeves, 2026 OK 37", "filing_status": "sanctioned", "jurisdiction": "Supreme Court of Oklahoma", "filing_date": "2026-05-27", "error_date": "2025", "public_record_link": "https://www.oscn.net/applications/oscn/DeliverDocument.asp?CiteID=551746", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.alnd.179677/gov.uscourts.alnd.179677.204.0.pdf", "confidence_score": "high", "notes_on_resolution": "Public reprimand (reciprocal; also reprimanded/censured in TX, TN, AL federal court). ChatGPT named and admitted by respondent. Underlying Johnson v. Dunn sanctions order, ECF No. 204, filed 2025-07-23, independently verified against the RECAP PDF.", "tags": "hallucinated-citations; attorney-discipline; chatgpt; reciprocal-discipline; legal-services; sanctions", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "neutral_citation": "2026 OK 37", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-022", "error_title": "Pro se appellant sanctioned for DeepSeek-hallucinated case citations", "ai_system_name": "DeepSeek (per appellant admission)", "deployer": "Pro se appellant", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "An Ohio trial court sanctioned pro se litigant Xingkui Guo $100 (2025-08-11) for citing five fabricated/mis-cited cases that the court found \"appear to be the result of AI-generated hallucinations.\" On appeal, Guo admitted the mis-citations resulted from his use of the Chinese AI platform \"Deepseek.\" The Sixth District Court of Appeals affirmed on 2026-05-26 (Guo v. Meade Motorcars, L.L.C., 2026-Ohio-1930), rejecting Guo’s motion to set aside the sanction.", "canonical_source_conflicted": "Genuine Ohio case law (cases cited did not exist at the given citations)", "mitigation_gap": "No verification of AI-generated authorities before filing.", "reliance_or_harm": "Fabricated citations submitted on appeal; $100 monetary sanction imposed.", "public_matter_type": "sanctions order", "public_matter_name": "Guo v. Meade Motorcars, L.L.C., 2026-Ohio-1930 (No. S-25-035)", "filing_status": "sanctioned", "jurisdiction": "Ohio Court of Appeals, Sixth Appellate District", "filing_date": "2026-05-26", "error_date": "2025", "public_record_link": "https://supremecourt.ohio.gov/rod/docs/pdf/6/2026/2026-Ohio-1930.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$100 sanction imposed by Sandusky County Court of Common Pleas 2025-08-11; Sixth District Court of Appeals affirmed denial of Guo’s motion to set aside, 2026-05-26. DeepSeek named by appellant; court found citations were AI-generated hallucinations.", "tags": "hallucinated-citations; pro-se; deepseek; legal-services; monetary-sanction; appellate-affirmance", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "neutral_citation": "2026-Ohio-1930", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-023", "error_title": "Attorney removed and denied fees over Westlaw CoCounsel-fabricated quotations", "ai_system_name": "Westlaw CoCounsel", "deployer": "Steven N. Howe (court-appointed appellate counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "In a published per curiam opinion, the Sixth Circuit found that attorney Steven N. Howe used Westlaw CoCounsel to draft appellate briefs and filed them without verifying the authorities. The briefs contained three fabricated quotations attributed to real cases and misrepresented the holdings of United States v. Washington and United States v. Anthony. Howe admitted the AI-generated errors.", "canonical_source_conflicted": "Genuine holdings of the cited authorities and the duty of candor to the tribunal", "mitigation_gap": "Relied on non-attorney \"staff\" to supervise AI output; no verification of citations before filing.", "reliance_or_harm": "False quotations and misleading legal arguments submitted in a criminal appeal; judicial resources expended; indigent defendant’s appeal delayed.", "public_matter_type": "appellate opinion / attorney sanction", "public_matter_name": "United States v. Farris, No. 25-5623 (6th Cir.)", "filing_status": "sanctioned", "jurisdiction": "U.S. Court of Appeals for the Sixth Circuit", "filing_date": "2026-04-03", "error_date": "2026", "public_record_link": "https://www.opn.ca6.uscourts.gov/opinions.pdf/26a0105p-06.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Recommended for publication. Howe denied CJA compensation for the appeal, removed as counsel, briefs locked, and referred to the Sixth Circuit Chief Judge (Local Rule 46) and the Kentucky Bar. Notably the fabrications were false quotations from REAL cases, not fictitious cases.", "tags": "hallucinated-citations; attorney-discipline; westlaw-cocounsel; legal-services; criminal-appeal", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-024", "error_title": "Supervising attorney sanctioned over associate’s AI fake citation (CoCounsel)", "ai_system_name": "Westlaw CoCounsel", "deployer": "Webb Law Group, APC (Lenden Webb, supervising attorney)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A magistrate judge sanctioned supervising attorney Lenden Webb after his firm filed a discovery letter brief containing a nonexistent case (South Pointe Wholesale v. Vilardi) and a misrepresented holding, \"caused in part by use of AI.\" The associate had used Westlaw CoCounsel. Webb was admonished and personally sanctioned for failure to supervise and failure to obey court orders.", "canonical_source_conflicted": "Genuine case law; duties of candor and supervision (Cal. Rules of Prof. Conduct 5.1)", "mitigation_gap": "No supervision of associate’s AI use; firm relied on a vendor representation that CoCounsel could not produce hallucinations.", "reliance_or_harm": "Fake citation submitted to the court; judicial resources expended across multiple show-cause orders.", "public_matter_type": "sanctions order", "public_matter_name": "Hill v. Workday, Inc., No. 3:23-cv-06558-PHK", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of California", "filing_date": "2026-04-28", "error_date": "2025", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.cand.422617/gov.uscourts.cand.422617.230.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$1,001 personal sanction + 4 hours CLE + circulation/certification requirements; associate Katherine Cervantes was separately sanctioned (Dkt. 161). Caption shares the name \"Workday\" with AIEL-2025-018 (Mobley) but is unrelated to the hiring-tool matter — here Workday is merely the opposing party and the AI misconduct is by plaintiff’s counsel.", "tags": "hallucinated-citations; attorney-discipline; supervision; westlaw-cocounsel; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-025", "error_title": "Plaintiff counsel sanctioned $2,500 for LexisNexis Protégé-hallucinated citations", "ai_system_name": "LexisNexis+ (Protégé)", "deployer": "Gavril T. Gabriel (plaintiff’s counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A district judge sanctioned plaintiff’s counsel Gavril T. Gabriel for citing a fabricated case (\"In re Shubert, 799 F.3d 1124\") three times across summary-judgment oppositions. Gabriel admitted using LexisNexis+ and its drafting feature Protégé. The court found bad faith, compounded by attempts to bury and minimize the misconduct.", "canonical_source_conflicted": "Genuine case law", "mitigation_gap": "No verification of AI-generated citations; concealment of AI use until ordered to disclose.", "reliance_or_harm": "Fabricated authority cited three times; bad-faith conduct found.", "public_matter_type": "sanctions order", "public_matter_name": "Chaney v. Transdev Services Inc., No. 2:24-cv-10761-ODW (AJRx)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Central District of California", "filing_date": "2026-04-28", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.cacd.951252/gov.uscourts.cacd.951252.46.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$2,500 total ($500 + $1,000 + $1,000) for bad faith; ordered to file a disclosure declaration in every Central District of California case in which he appears.", "tags": "hallucinated-citations; attorney-discipline; lexisnexis-protege; bad-faith; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-026", "error_title": "S.D.N.Y. sanctions attorney $2,500 for AI-fabricated citations; LexisNexis attribution rejected", "ai_system_name": "Unspecified generative-AI program (court finding; attorney’s LexisNexis attribution not credited)", "deployer": "Tricia S. Lindsay (plaintiff’s counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A magistrate judge sanctioned attorney Tricia S. Lindsay after she filed two opposition memoranda citing numerous nonexistent cases. By process of elimination (no human drafter claimed), the court found the citations \"were generated as a result of the use of some kind of AI program,\" expressly declining to credit Lindsay’s belated suggestion that LexisNexis was the source. The court found subjective bad faith.", "canonical_source_conflicted": "Genuine case law; Rule 11 duty of reasonable inquiry", "mitigation_gap": "No citation verification; obfuscation about the source of the fabricated citations.", "reliance_or_harm": "$2,500 Rule 11 monetary sanction against plaintiff’s counsel; numerous fabricated citations across multiple filings; counsel referred, with co-counsel, to the Second Circuit Grievance Panel.", "public_matter_type": "sanctions opinion (Rule 11)", "public_matter_name": "Jimenez-Fogarty v. Fogarty, No. 1:24-cv-08705-JLR-GWG (S.D.N.Y.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2026-04-29", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.631979/gov.uscourts.nysd.631979.192.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$2,500 Rule 11 sanction; ordered to notify the judge in every pending case where she appears; separately referred (with co-counsel Jo Saint-George) to the Second Circuit Grievance Panel per Ioio v. City of New York (2d Cir. Mar. 16, 2026). NOTE: the Charlotin \"LexisNexis\" tool tag is NOT supported — the court declined to credit LexisNexis as the source. Merged 2026-07-24 with the duplicate record AIEL-2026-039, which described the same order in the same docket on the same date; AIEL-2026-039 is retired and must not be reused.", "tags": "hallucinated-citations; attorney-discipline; unspecified-ai; bad-faith; legal-services; court-finding; unnamed-ai; rejected-tool-attribution", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "domain": "legal services", "error_type": "hallucinated authority", "error_date": "2026", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "secondary_source_links": "https://publicportal.alappeals.gov/portal/court/68f021c4-6a44-4735-9a76-5360b2e8af13/case/47683afc-2f85-469d-8562-9dca094bb680", "confidence_score": "high", "error_id": "AIEL-2026-027", "error_title": "Alabama Supreme Court sanctions attorney for AI-fabricated authorities", "ai_system_name": "Unnamed generative-AI research tool (court finding; counsel conceded)", "deployer": "Appellants’ counsel (attorney \"Hall\")", "error_description": "The Supreme Court of Alabama found that briefs filed by appellants’ counsel (\"Hall\") contained numerous invalid citations the court determined were \"AI hallucinations,\" i.e., fake authorities created by an AI system. Counsel conceded the improper AI use, describing it as his first use of an AI research tool. The court dismissed the appeal as grossly deficient and imposed sanctions.", "canonical_source_conflicted": "Genuine legal authorities; Alabama Rules of Appellate Procedure", "mitigation_gap": "No verification of AI-generated authorities before filing.", "reliance_or_harm": "Fabricated authorities forced opposing party and the court to expend resources; appeal dismissed.", "public_matter_type": "appellate sanction / dismissal", "public_matter_name": "Ibach v. Stewart, No. SC-2025-0106 (Ala.)", "filing_status": "sanctioned", "jurisdiction": "Supreme Court of Alabama", "filing_date": "2026-04-24", "public_record_link": "https://publicportal-api.alappeals.gov/courts/68f021c4-6a44-4735-9a76-5360b2e8af13/cms/case/47683afc-2f85-469d-8562-9dca094bb680/docketentrydocuments/d95cdd15-7428-4cc9-8389-3cc27ac88089", "notes_on_resolution": "Counsel ordered to pay $17,200 in attorney fees and costs to the appellee, plus double costs, and barred from further filings absent leave; appeal dismissed.", "tags": "hallucinated-citations; attorney-discipline; unnamed-ai; legal-services; appellate", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "domain": "legal services", "error_type": "hallucinated authority", "error_date": "2026", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.paed.609239/gov.uscourts.paed.609239.226.0.pdf", "confidence_score": "high", "error_id": "AIEL-2026-028", "error_title": "Attorney sanctioned $5,000 for repeated improper AI use (second offense)", "ai_system_name": "Unnamed generative AI (ChatGPT referenced in order)", "deployer": "Defendants’ counsel Raja Rajan", "error_description": "A district judge sanctioned defendants’ attorney Raja Rajan $5,000 under Rule 11(c) for his repeated improper use of generative AI in filings — his second such violation. The court rejected his plea for a lighter penalty, noting prior sanctions exceeding $73,000.", "canonical_source_conflicted": "Genuine case law; Rule 11 duty of inquiry", "mitigation_gap": "Failed to verify AI output despite a prior sanction and the court’s standing order on AI use.", "reliance_or_harm": "Repeat submission of unverified AI-derived filings; escalating sanctions.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Bunce v. Visual Technology Innovations, Inc., No. 2:23-cv-01740-KNS", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Eastern District of Pennsylvania", "filing_date": "2026-04-20", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.paed.609239/gov.uscourts.paed.609239.225.0.pdf", "notes_on_resolution": "$5,000 Rule 11(c) sanction plus an AI-ethics CLE requirement; second AI-related sanction against the same attorney (prior sanctions exceeded $73,000).", "tags": "hallucinated-citations; attorney-discipline; repeat-offender; generative-ai; legal-services", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "domain": "legal services", "error_type": "hallucinated authority", "error_date": "2026", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.ilnd.475370/gov.uscourts.ilnd.475370.96.0_1.pdf", "confidence_score": "high", "error_id": "AIEL-2026-029", "error_title": "Pro se plaintiff sanctioned $5,000; court found brief generated using AI", "ai_system_name": "Unnamed generative AI (court finding)", "deployer": "Pro se plaintiff Ifeoma Delliane Chinedu Obi", "error_description": "A district court struck a pro se plaintiff’s Rule 59(e) motion and sanctioned her $5,000 under Rule 11 after finding she \"generated each brief using AI.\" The filings contained at least 13 hallucinated cases, quotations, and statements of law and fact.", "canonical_source_conflicted": "Genuine case law; Rule 11", "mitigation_gap": "No verification of AI-generated content; repeated violations after notice.", "reliance_or_harm": "Multiple filings riddled with fabricated authority; motion struck.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Obi v. Cook County, Illinois, No. 1:25-cv-03096", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Illinois", "filing_date": "2026-04-09", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.ilnd.475370/gov.uscourts.ilnd.475370.97.0.pdf", "notes_on_resolution": "Motion struck (Local Rule 7.1) and $5,000 Rule 11 sanction. NOTE: Charlotin reported $4,999; the order imposes $5,000. Pro se status did not shelter the plaintiff from sanctions.", "tags": "hallucinated-citations; pro-se; unnamed-ai; legal-services; monetary-sanction", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "domain": "legal services", "error_type": "hallucinated authority", "error_date": "2026", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "secondary_source_links": "", "confidence_score": "high", "error_id": "AIEL-2026-030", "error_title": "Pro se appellant sanctioned $3,178 for AI-produced fictitious authorities", "ai_system_name": "Unnamed generative legal-research tool (per appellant; court finding)", "deployer": "Pro se plaintiff-appellant Robert S. Hulvat", "error_description": "An Illinois appellate court found that pro se appellant Robert S. Hulvat filed a brief \"produced using artificial intelligence\" that cited nine nonexistent cases and fictitious holdings, violating the Illinois Supreme Court AI policy and appellate rules. Hulvat asserted he had used a \"well-known generative legal research tool.\"", "canonical_source_conflicted": "Genuine Illinois case law; Illinois Supreme Court Policy on AI; Rule 375", "mitigation_gap": "No verification of AI-generated authorities before filing.", "reliance_or_harm": "Fictitious authorities forced opposing counsel to expend fees; dismissal affirmed.", "public_matter_type": "appellate sanction", "public_matter_name": "Hulvat v. Gumina, 2026 IL App (3d) 240628-U", "filing_status": "sanctioned", "jurisdiction": "Appellate Court of Illinois, Third District", "filing_date": "2026-04-09", "public_record_link": "https://illinoiscourts.gov/resources/63539016-cb02-4575-b3da-0dd0632244c1/file", "notes_on_resolution": "Judgment for fees and costs of $3,178 entered against the appellant under Illinois Supreme Court Rule 375(b); dismissal of the complaint affirmed.", "tags": "hallucinated-citations; pro-se; unnamed-ai; legal-services; appellate", "neutral_citation": "2026 IL App (3d) 240628-U", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "domain": "legal services", "error_type": "hallucinated authority", "error_date": "2026", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "secondary_source_links": "", "confidence_score": "high", "error_id": "AIEL-2026-031", "error_title": "Bankruptcy attorney sanctioned $2,750 + AI CLE after admitting AI use", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Debtor’s counsel Kathleen Wilson", "error_description": "A bankruptcy court sanctioned attorney Kathleen Wilson after she admitted using generative AI to prepare a Supplemental Response and other pleadings without verifying the output. The filings contained AI-generated citations to nonexistent cases, including In re Garcia and In re Buttermilk Towne LLC.", "canonical_source_conflicted": "Genuine case law; Fed. R. Bankr. P. 9011", "mitigation_gap": "Failed to review or verify admittedly AI-generated content before filing.", "reliance_or_harm": "False citations submitted to the court; sanctions and mandatory AI education imposed.", "public_matter_type": "sanctions order (Rule 9011)", "public_matter_name": "In re Wise, No. 25-51132 (Bankr. W.D. La.)", "filing_status": "sanctioned", "jurisdiction": "U.S. Bankruptcy Court, Western District of Louisiana", "filing_date": "2026-04-09", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.lawb.374130/gov.uscourts.lawb.374130.84.0.pdf", "notes_on_resolution": "$2,750 monetary sanction payable to the Clerk plus six hours of in-person CLE focused on the use of generative AI; counsel admitted the AI use.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; bankruptcy; legal-services", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-032", "error_title": "California appellate counsel sanctioned $5,000 for AI-hallucinated citations", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Roxanne Chung Bonar (appellate counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The California Court of Appeal sanctioned respondent's appellate counsel Roxanne Chung Bonar $5,000 after she cited fictitious cases in a family-law appeal. At oral argument Bonar acknowledged she lacked a paid legal-research subscription and conceded the fictitious citations \"may have\" come from AI tools she was using. The same fake cases had also appeared in the family court's written order, which the appellate court found was error.", "canonical_source_conflicted": "Non-existent California authorities (Twigg and Teegarden); genuine California family-law caselaw", "mitigation_gap": "Counsel used AI for legal research without a verified subscription and submitted citations without checking their existence.", "reliance_or_harm": "$5,000 sanction payable to court clerk; referral to State Bar of California; family court order reversed for reliance on fictitious authority.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "Torres Campos v. Munoz, No. D085584 (Cal. Ct. App., 4th Dist., Div. One)", "filing_status": "sanctioned", "jurisdiction": "California Court of Appeal, Fourth Appellate District, Division One", "filing_date": "2026-03-05", "error_date": "2026", "public_record_link": "https://courts.ca.gov/opinions/archive/D085584.PDF", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$5,000 sanction and bar referral confirmed; AI unnamed but admitted at oral argument.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; family-law; legal-services; bar-referral", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-033", "error_title": "Louisiana attorney sanctioned $1,000 for ChatGPT-fabricated citations", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Kenneth A. Mayeaux (Mayeaux & Associates)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Kenneth A. Mayeaux used ChatGPT to draft a memorandum in support of a habeas petition and submitted it without verifying the citations. The memorandum contained hallucinated case law. Mayeaux admitted both using ChatGPT and failing to vet the citations before filing. The court sanctioned him $1,000 personally under Fed. R. Civ. P. 11 and 28 U.S.C. § 1927.", "canonical_source_conflicted": "Genuine federal immigration/habeas caselaw; Fed. R. Civ. P. 11", "mitigation_gap": "Attorney used ChatGPT to draft legal memoranda and filed without any citation verification.", "reliance_or_harm": "$1,000 personal sanction against counsel; false citations submitted in habeas proceedings.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Kattom v. Bondi, No. 25-1497 SEC P (W.D. La.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Western District of Louisiana", "filing_date": "2026-03-06", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.lawd.213966/gov.uscourts.lawd.213966.28.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$1,000 personal sanction (2026 WL 637419); attorney admitted ChatGPT use and failure to verify.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; habeas; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-034", "error_title": "E.D. Louisiana attorneys sanctioned $1,250 for ChatGPT-hallucinated citations", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Jalen R. Harris; James Roquemore; Corwin St. Raymond", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Drafting attorney Jalen R. Harris admitted starting a brief on Westlaw then \"resorting to ChatGPT,\" submitting nine hallucinated case citations without checking them. The court imposed $250 on Harris and $1,000 on supervising attorney James Roquemore under Rule 11, and formally admonished co-counsel Corwin St. Raymond. Harris took full responsibility and acknowledged he never read the fabricated cases.", "canonical_source_conflicted": "Nine non-existent federal cases cited in the brief; genuine federal law", "mitigation_gap": "Drafting attorney used ChatGPT without verifying output; supervising attorney failed to catch the hallucinations.", "reliance_or_harm": "$1,250 in monetary sanctions ($250 + $1,000); formal admonishment of co-counsel.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Gentry v. Thompson, No. 2:25-cv-01260-CJB-EJD (E.D. La.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Eastern District of Louisiana", "filing_date": "2026-03-20", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.laed.273860/gov.uscourts.laed.273860.25.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$250 Harris + $1,000 Roquemore; St. Raymond admonished only; Harris fully admitted ChatGPT use.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-035", "error_title": "Ohio appellate attorney sanctioned $2,000 for ChatGPT-fabricated quotes, continues AI use post-hearing", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Respondent attorney (Ohio appellate counsel; name not stated in order)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The Ohio Court of Appeals, Eleventh Appellate District, sanctioned an attorney after a non-attorney paralegal used ChatGPT to generate fabricated quotations that the attorney submitted in a criminal appeal. The attorney admitted the AI use and the court found the conduct sanctionable; it also noted the attorney had continued using ChatGPT after the sanctions hearing. Aggravating: a subsequent October 2025 filing visibly contained a ChatGPT prompt.", "canonical_source_conflicted": "Non-existent judicial quotations attributed to real Ohio judges; Ohio Rules of Professional Conduct", "mitigation_gap": "Attorney failed to review paralegal's ChatGPT output before submission; continued AI use after sanctions hearing.", "reliance_or_harm": "$2,000 monetary fine (credit for $2,000 already paid to Ashtabula County Prosecutor); referral to Ohio Office of Disciplinary Counsel; removal as counsel; application to reopen stricken; two-year order to serve judgment on falsely-implicated judges.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "State v. Coleman, 2026-Ohio-965, No. 2024-A-0040 (Ohio Ct. App., 11th Dist.)", "filing_status": "sanctioned", "jurisdiction": "Court of Appeals of Ohio, Eleventh Appellate District", "filing_date": "2026-03-20", "error_date": "2026", "public_record_link": "https://supremecourt.ohio.gov/rod/docs/pdf/11/2026/2026-Ohio-965.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$2,000 fine with $2,000 credit for prior Ashtabula County payment; disciplinary referral; removal; appeal stricken.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; criminal-appeal; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "neutral_citation": "2026-Ohio-965", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-036", "error_title": "D. Oregon attorney sanctioned $15,500 + $94,704 fees for AI-hallucinated briefs", "ai_system_name": "Unnamed generative AI (court finding)", "deployer": "Stephen Brigandi (pro hac vice lead counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "In a two-phase sanctions proceeding, the U.S. District Court for the District of Oregon struck plaintiff's summary-judgment briefs, dismissed claims with prejudice, and imposed sanctions on pro hac vice counsel Stephen Brigandi for submitting briefing containing non-existent cases and fabricated quotations that the court attributed to an unidentified AI tool. A December 2025 order imposed a $15,500 Rule 11 sanction; the March 2026 order finalized a fee award of $94,704.38 apportioned 85% to Brigandi and 15% to local counsel Timothy Murphy.", "canonical_source_conflicted": "Non-existent case citations and fabricated quotations; Fed. R. Civ. P. 11", "mitigation_gap": "Lead counsel submitted AI-generated briefing without verification; court found \"one of the clients may have played a role in generating the fake citations\" via an unidentified AI tool.", "reliance_or_harm": "$15,500 monetary sanction on Brigandi; $94,704.38 attorney-fee award ($80,498.72 Brigandi, $14,205.66 Murphy); claims dismissed with prejudice; summary-judgment briefs stricken.", "public_matter_type": "sanctions order (Rule 11 / inherent power)", "public_matter_name": "Couvrette v. Wisnovsky, No. 1:21-cv-00157-CL (D. Or.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Oregon", "filing_date": "2026-03-23", "error_date": "2025", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.ord.158388/gov.uscourts.ord.158388.225.0.pdf", "secondary_source_links": "https://courtlistener.com/docket/55220104/226/couvrette-v-wisnovsky/", "confidence_score": "high", "notes_on_resolution": "Two-order matter: Dec 12, 2025 (AI sanction + $15,500 + claims dismissed); March 23, 2026 (fee split finalized). Total exposure ~$110,204. On March 24, 2026, ECF No. 226 recorded a Rule 60(a) administrative correction because an incorrect PDF had been attached to ECF No. 225. The court replaced that attachment with a corrected PDF for scrivener's errors. The corrected ECF No. 225 retains the $94,704.38 total and exact 85/15 allocation ($80,498.72 to Brigandi and $14,205.66 to Murphy), so the corpus figures are unchanged.", "tags": "hallucinated-citations; attorney-discipline; unnamed-ai; court-finding; rule-11; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-25", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-037", "error_title": "S.D. Alabama attorney sanctioned $55,597 for CoCounsel-generated gibberish filings", "ai_system_name": "CoCounsel (Westlaw) (named; court-found; counsel's denial not credited)", "deployer": "Franklin Hollis Eaton, Jr. (plaintiff's counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The Southern District of Alabama publicly reprimanded attorney Franklin Hollis Eaton, Jr. and imposed a $55,597 sanction after he submitted hundreds of pages of gibberish produced directly from Westlaw's CoCounsel AI service. Eaton denied using AI, but the court found his denial \"incredible\" because he had himself attached CoCounsel-generated documents. The order was directed published in the Federal Supplement and referred to the Alabama State Bar.", "canonical_source_conflicted": "Accurate pleading standards; Court Local Rules; Fed. R. Civ. P. 11", "mitigation_gap": "Counsel attached raw, unreviewed CoCounsel output containing hundreds of pages of erroneous information, then denied AI use.", "reliance_or_harm": "$55,597 attorney-fee sanction; published public reprimand; Bar referral; order-to-file-in-subsequent-cases for 12 months.", "public_matter_type": "sanctions order (Rule 11 / inherent power)", "public_matter_name": "Heimkes v. Fairhope Motorcoach Resort Condominium Owners Association et al., No. 1:22-cv-00448-TFM-N (S.D. Ala.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of Alabama", "filing_date": "2026-03-31", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.alsd.70919/gov.uscourts.alsd.70919.353.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$55,597.00 confirmed; counsel denied AI use but court expressly found he \"clearly used it at some point\"; denial not credited.", "tags": "hallucinated-citations; attorney-discipline; court-finding; cocounsel; westlaw; denied-ai; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-038", "error_title": "N.D. Alabama attorney disqualified and sanctioned $47,057 for ChatGPT-hallucinated citations", "ai_system_name": "ChatGPT (named; court-found)", "deployer": "Joshua B. Watkins (Burrill Watkins LLC)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Joshua B. Watkins of Burrill Watkins LLC was publicly reprimanded, disqualified from the case, and sanctioned a combined $47,056.90 after forensic examination of his firm's ChatGPT Business Plan workspace confirmed hallucinated citations in filings. Counsel admitted generative-AI use and the court made an independent finding of false statements of law. The order was published in the Federal Supplement; the Bar was notified.", "canonical_source_conflicted": "Non-existent federal case citations; genuine federal civil rights and landlord-tenant law", "mitigation_gap": "Attorney used ChatGPT to draft legal briefs and filed without verifying citations; forensic examination of the firm's ChatGPT workspace confirmed the AI origin.", "reliance_or_harm": "$35,603.90 to Fite Defendants + $11,453 to Triad Defendants = $47,056.90; public reprimand; disqualification; Bar referral; order published.", "public_matter_type": "sanctions order (Rule 11 / inherent power)", "public_matter_name": "Rivera v. Triad Properties Corporation et al., No. 2:24-cv-01802-AMM (N.D. Ala.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Alabama", "filing_date": "2026-03-31", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.alnd.192599/gov.uscourts.alnd.192599.116.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$47,056.90 total ($35,603.90 + $11,453); ChatGPT workspace forensically examined; both admission and court finding confirmed.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; disqualification; bar-referral; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-040", "error_title": "D. Colorado attorney sanctioned $5,000 in second AI-hallucination order (Coomer v. MyPillow)", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Christopher Kachouroff", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "A second Rule 11 sanctions order in the Coomer v. Lindell/MyPillow litigation imposed a $5,000 sanction on attorney Christopher Kachouroff, who had earlier admitted to routinely using AI in drafting and whose prior brief contained approximately 30 defective or hallucinated AI-generated citations. The court also referred Kachouroff to the Colorado State Bar. Cumulative sanctions in the matter reached $8,000.", "canonical_source_conflicted": "Accurate legal authority; Fed. R. Civ. P. 11", "mitigation_gap": "Counsel admitted routine AI use in drafting; failed to verify citations; second violation in same litigation.", "reliance_or_harm": "$5,000 monetary sanction (second order); Bar referral; cumulative $8,000 in sanctions in this matter.", "public_matter_type": "sanctions order (Rule 11) — second order", "public_matter_name": "Coomer v. Lindell/MyPillow Inc., No. 1:22-cv-01129-NYW-SBP (D. Colo.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Colorado", "filing_date": "2026-05-07", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.cod.215068/gov.uscourts.cod.215068.424.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Second standalone sanctions order ($5,000); first order imposed $3,000 on Kachouroff + $3,000 on co-counsel DeMaster. Bar referral confirmed.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; repeat-offense; bar-referral; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-041", "error_title": "S.D.N.Y. defense counsel sanctioned $2,710 for LexisNexis AI-hallucinated citations", "ai_system_name": "LexisNexis AI tool (named; admitted)", "deployer": "Defense Counsel for Sushi Katsuei (unnamed in opinion)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Defense counsel in a wage-and-hour class action admitted using LexisNexis's AI tool to help construct submissions without reviewing them for correctness, resulting in AI-hallucinated citations. The court imposed $1,710 in attorneys' fees (AI-authority-related) and a $1,000 monetary penalty on defense counsel under Rule 37, 28 U.S.C. § 1927, and inherent powers. A separate $46,980 discovery sanction was also imposed on defendants for unrelated witness-declaration misconduct.", "canonical_source_conflicted": "Genuine federal employment/wage-and-hour caselaw; court submission standards", "mitigation_gap": "Counsel used LexisNexis AI to draft court submissions without reviewing AI output for accuracy before filing.", "reliance_or_harm": "$1,710 AI-attributable attorneys' fees + $1,000 monetary penalty = $2,710 on defense counsel; fabricated citations submitted in class-action proceedings.", "public_matter_type": "sanctions order (Rule 37 / inherent power)", "public_matter_name": "Chakma v. Sushi Katsuei Inc., No. 1:23-cv-07804-KPF (S.D.N.Y.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2026-05-19", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.605627/gov.uscourts.nysd.605627.142.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "AI-attributable sanction $2,710 ($1,710 fees + $1,000 penalty); LexisNexis AI confirmed by counsel admission and court finding. Separate $46,980 discovery sanction not AI-related.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; court-finding; lexisnexis-ai; wage-and-hour; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-042", "error_title": "S.D. Indiana attorney sanctioned $2,000 for Fastcase AI-hallucinated citations", "ai_system_name": "Fastcase (named; court-found)", "deployer": "Tae Sture (plaintiff's counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The Southern District of Indiana sanctioned attorney Tae Sture $2,000 under Rule 11 after his firm's paralegal used the Fastcase legal research tool, which generated hallucinated citations that were filed without verification. The court characterized the citations as \"hallucinations generated by an artificial intelligence research tool\" and specifically named Fastcase. The district judge reduced the Magistrate's recommended $7,500 sanction to $2,000.", "canonical_source_conflicted": "Genuine federal employment and civil-rights caselaw; Fed. R. Civ. P. 11", "mitigation_gap": "Paralegal used Fastcase AI without verification; supervising attorney failed to review citations before filing.", "reliance_or_harm": "$2,000 Rule 11 sanction (reduced from $7,500 recommended); AI-hallucinated citations submitted in federal employment litigation.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Davis v. Marion County Superior Court Juvenile Detention Center, No. 1:24-cv-01918-JRO-MJD (S.D. Ind.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of Indiana", "filing_date": "2026-05-20", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.insd.217901/gov.uscourts.insd.217901.127.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$2,000 sanction (reduced from magistrate's $7,500 recommendation); Fastcase named as the AI tool by court finding.", "tags": "hallucinated-citations; attorney-discipline; court-finding; fastcase; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-043", "error_title": "7th Circuit sanctions attorney $5,000 for ChatGPT-hallucinated brief with 24 fabricated quotations", "ai_system_name": "ChatGPT (named; court-found)", "deployer": "Abdullah Salah (counsel of record); Farah Chalisa (contracted brief-drafter)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The Seventh Circuit found that an immigration appeal brief was \"riddled with AI hallucinations\" including two dozen fabricated quotations and seven cases wrongly cited as circuit precedent. The brief was drafted by contracted attorney Farah Chalisa using ChatGPT, and filed by counsel of record Abdullah Salah. Salah was sanctioned $5,000 under Fed. R. App. P. 46(c); Chalisa was referred to the Illinois ARDC but not directly sanctioned because she had not appeared before the circuit.", "canonical_source_conflicted": "Seventh Circuit immigration precedent; Fed. R. App. P. 46(c)", "mitigation_gap": "Contracted drafter used ChatGPT for copyediting without verification; counsel of record signed and filed without reviewing for AI hallucinations.", "reliance_or_harm": "$5,000 sanction on Salah; ARDC referral for Chalisa; court placed Salah on notice of potential disbarment; petition denied.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "Perez-Castillo v. Blanche (Att'y Gen.), No. 25-1988 (7th Cir.)", "filing_status": "sanctioned", "jurisdiction": "U.S. Court of Appeals for the Seventh Circuit", "filing_date": "2026-06-01", "error_date": "2026", "public_record_link": "https://media.ca7.uscourts.gov/cgi-bin/OpinionsWeb/processWebInputExternal.pl?Submit=Display&Path=Y2026/D06-01/C:25-1988:J:Brennan:aut:T:fnOp:N:3550588:S:0", "secondary_source_links": "https://govinfo.gov/content/pkg/USCOURTS-ca7-25-01988/pdf/USCOURTS-ca7-25-01988-0.pdf", "confidence_score": "high", "notes_on_resolution": "$5,000 on Salah (not $4,997 as in tracker); ARDC referral for Chalisa; 24 fabricated quotations and 7 wrong-circuit citations documented.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; court-finding; chatgpt; immigration; seventh-circuit; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-044", "error_title": "NY Appellate Division sanctions attorney + client $10,000 for 23 AI-fabricated case citations", "ai_system_name": "Unnamed generative AI (court finding; filer denied AI use)", "deployer": "Joshua A. Douglass (defense appellate counsel) and Jean LeTennier (defendant/client)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "The New York Appellate Division, Third Department, found that defendant's five appellate filings contained at least 23 fabricated case citations. Document metadata showed the filings originated from a program in the client's name. Despite the defendant denying AI use, the court made an express, unhedged finding of generative-AI misuse — the first reported GenAI sanctions decision at the New York appellate level. Sanctions of $7,500 were imposed on counsel and $2,500 on the client.", "canonical_source_conflicted": "23 non-existent case citations; 22 NYCRR 130-1.1 (frivolous conduct)", "mitigation_gap": "Neither counsel nor client verified AI-generated citations before filing; client apparently ran the AI tool and counsel signed submissions without review.", "reliance_or_harm": "$10,000 in sanctions ($7,500 on attorney Douglass + $2,500 on LeTennier); appeals affirmed with costs.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "Deutsche Bank Natl. Trust Co. v. LeTennier, 2026 NY Slip Op 00040, Docket No. CV-23-0713 (N.Y. App. Div., 3d Dep’t)", "filing_status": "sanctioned", "jurisdiction": "New York Supreme Court, Appellate Division, Third Department", "filing_date": "2026-01-08", "error_date": "2026", "public_record_link": "https://nycourts.gov/reporter/3dseries/2026/2026_00040.htm", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$7,500 on counsel + $2,500 on client = $10,000; metadata showed filings originated from client's AI program; first NY appellate-level GenAI sanctions decision. Source is the New York Official Reports slip opinion (official reporter, HTML). Tooling note: nycourts.gov returns HTTP 403 to bare scripted requests but 200 to a browser-like User-Agent, so an automated link-check failure here is a known false negative; both the bare and www hosts serve the document.", "tags": "hallucinated-citations; court-finding; unnamed-ai; appellate; attorney-discipline; client-sanctioned; new-york; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-045", "error_title": "M.D. Florida counsel sanctioned $7,000 in two companion cases for AI-fabricated citations", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Plaintiff's counsel (name not in order)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Plaintiff's counsel admitted filing briefs in two companion employment cases containing AI-generated fabricated citations. The U.S. District Court for the Middle District of Florida imposed a $7,000 fee sanction and required counsel to deliver ten speaking engagements at bar associations or law-school groups by January 2027 to educate the profession on AI risks. The court declined to refer counsel to the Florida Bar or the court's Grievance Committee.", "canonical_source_conflicted": "Non-existent federal case citations filed in two employment-law matters; Fed. R. Civ. P. 11", "mitigation_gap": "Counsel admitted reliance on AI-generated research without verifying citation existence before filing in both cases.", "reliance_or_harm": "$7,000 fee sanction plus mandatory 10 bar-association/law-school speaking engagements by January 2027.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Hodges v. Meridian Waste Acquisitions USA, No. 3:25-cv-62; Rieske v. AFS, No. 5:25-cv-45 (M.D. Fla.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Middle District of Florida", "filing_date": "2026-01-16", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.flmd.436812/gov.uscourts.flmd.436812.53.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Companion cases in single order; $7,000 fee sanction plus 10 mandatory speaking engagements; no bar referral.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; companion-cases; employment; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-046", "error_title": "E.D. Pennsylvania attorney sanctioned $4,000 for Lexis+ AI / LexisNexis Protégé hallucinated citations", "ai_system_name": "Lexis+ AI and LexisNexis Protégé (named; court-found via firm's sworn account)", "deployer": "Attorney Anderson (Anderson and Associates, specially admitted NY co-counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "An unnamed law clerk used Lexis+ AI and LexisNexis Protégé to research a filing, generating at least eight false case citation hallucinations. Supervising attorney Anderson signed the brief without verifying the AI output. The court found both co-counsel violated Rule 11 but imposed the $4,000 monetary sanction only on Anderson, crediting the firm's sworn account of the law clerk's AI tool use without rejecting the LexisNexis attribution.", "canonical_source_conflicted": "Eight non-existent federal case citations; Fed. R. Civ. P. 11", "mitigation_gap": "Supervising attorney failed to verify AI-generated research before signing and filing; AI detectors flagged the case explanations as 100% AI-generated.", "reliance_or_harm": "$4,000 Rule 11 sanction on supervising attorney; local co-counsel found to have violated Rule 11 but not fined.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Lifetime Well LLC v. IBSpot.com Inc., No. 2:25-cv-05135-MAK (E.D. Pa.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Eastern District of Pennsylvania", "filing_date": "2026-01-26", "error_date": "2026", "public_record_link": "https://govinfo.gov/content/pkg/USCOURTS-paed-2_25-cv-05135/pdf/USCOURTS-paed-2_25-cv-05135-0.pdf", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.paed.643304/gov.uscourts.paed.643304.42.0.pdf", "confidence_score": "high", "notes_on_resolution": "$4,000 on Anderson; local co-counsel Goldin also found to violate Rule 11 but not fined; LexisNexis tool attribution credited from firm's sworn show-cause response.", "tags": "hallucinated-citations; attorney-discipline; court-finding; lexisnexis-ai; lexis-protege; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-047", "error_title": "D. Kansas $12,000 sanctions on five attorneys for ChatGPT-fabricated citations (patent case)", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Sandeep Seth (drafting attorney); Kenneth P. Kula; Eric W. Joe; David R. Cooper (co-counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Sandeep Seth admitted using ChatGPT \"as a shortcut\" to find case law and submitted fabricated and misrepresented citations in a patent infringement matter without independent verification. The court found all five Lexos attorneys violated Rule 11(b). Seth was fined $5,000 and had his pro hac vice admission revoked; three co-counsel were fined $3,000, $3,000, and $1,000 respectively for failure to verify their co-counsel's work.", "canonical_source_conflicted": "Non-existent Tenth and Federal Circuit patent cases; Fed. R. Civ. P. 11(b)", "mitigation_gap": "Lead drafter used ChatGPT as research shortcut and admitted not verifying citations; four co-counsel signed filings without checking a co-counsel's citations.", "reliance_or_harm": "$12,000 total Rule 11 sanctions across four attorneys; Seth's pro hac vice admission revoked and required to self-report to his bar.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Lexos Media IP, LLC v. Overstock.com, Inc., No. 2:22-cv-02324-JAR (D. Kan.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Kansas", "filing_date": "2026-02-02", "error_date": "2025", "public_record_link": "https://govinfo.gov/content/pkg/USCOURTS-ksd-2_22-cv-02324/pdf/USCOURTS-ksd-2_22-cv-02324-9.pdf", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.ksd.142916/gov.uscourts.ksd.142916.218.0.pdf", "confidence_score": "high", "notes_on_resolution": "$12,000 aggregate: Seth $5,000 + pro hac vice revocation; Kula $3,000; Joe $3,000; Cooper $1,000; fifth attorney (Doell) released from sanctions.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; patent; multi-attorney; pro-hac-vice; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-048", "error_title": "10th Circuit sanctions Maryland attorney $1,000 for ChatGPT-hallucinated citations in her own appeal", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Kusmin L. Amarsingh (attorney appearing pro se on her own appeal)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Kusmin Amarsingh, appearing pro se in her own employment appeal against Frontier Airlines, admitted using ChatGPT as a research and drafting aid that generated seven fabricated case citations and two misattributed quotations. She admitted not verifying any of the citations. The Tenth Circuit affirmed the judgment against her, imposed a $1,000 sanction payable to Frontier, and referred the matter to Maryland attorney-disciplinary authorities.", "canonical_source_conflicted": "Seven non-existent federal cases plus two real cases cited with fabricated quotations; Fed. R. App. P.", "mitigation_gap": "Attorney used ChatGPT for research and drafting and filed without verifying any citations or quotations.", "reliance_or_harm": "$1,000 sanction payable to Frontier Airlines; referral to Maryland attorney-disciplinary authority; appeal affirmed against her.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "Amarsingh v. Frontier Airlines, Inc., No. 24-1391 (10th Cir.)", "filing_status": "sanctioned", "jurisdiction": "U.S. Court of Appeals for the Tenth Circuit", "filing_date": "2026-02-09", "error_date": "2025", "public_record_link": "https://ca10.uscourts.gov/sites/ca10/files/opinions/010111382504.pdf", "secondary_source_links": "https://govinfo.gov/content/pkg/USCOURTS-ca10-24-01391/pdf/USCOURTS-ca10-24-01391-0.pdf", "confidence_score": "high", "notes_on_resolution": "$1,000 to Frontier (fee offset); mandatory Maryland bar referral; attorney appeared pro se on her own employment appeal.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; chatgpt; pro-se-attorney; appellate; employment; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-049", "error_title": "California Court of Appeal sanctions attorney $900 for AI-hallucinated citations drafted by AI consultant", "ai_system_name": "Unnamed generative AI (court finding; AI \"consultant\" intermediary)", "deployer": "David E. Azar (EFD appellate counsel)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney David E. Azar commissioned an AI \"consultant\" to draft an appellate opening brief using an AI system, then filed the brief without adequately verifying citations despite being warned by the consultant to double-check for hallucinations. The California Court of Appeal found the fabricated citations were caused primarily by Azar's failure to safeguard against AI hallucinations and imposed a $900 sanction on him alone, exonerating co-counsel who only formatted and filed the document.", "canonical_source_conflicted": "Non-existent California appellate authorities cited in opening brief; Cal. Rules of Court", "mitigation_gap": "Attorney outsourced drafting to an AI consultant, ignored the consultant's own warning to verify citations, and filed without review.", "reliance_or_harm": "$900 sanction on Azar; judgment against EFD affirmed; co-counsel exonerated.", "public_matter_type": "sanctions order (appellate)", "public_matter_name": "EFD USA, Inc. v. Band Pro Film and Digital, Inc., No. B329314 (Cal. Ct. App., 2d Dist., Div. Three) (unpublished)", "filing_status": "sanctioned", "jurisdiction": "California Court of Appeal, Second Appellate District", "filing_date": "2026-02-18", "error_date": "2025", "public_record_link": "https://courts.ca.gov/opinions/nonpub/B329314.PDF", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "$900 on Azar alone; AI tool unnamed (accessed via unnamed AI \"consultant\"); co-counsel Keiter exonerated.", "tags": "hallucinated-citations; attorney-discipline; court-finding; unnamed-ai; ai-consultant; appellate; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-050", "error_title": "N.D. Indiana sanctions Joe Exotic's attorney $1,500 for AI-assisted fabricated citations", "ai_system_name": "Unnamed generative AI (admitted by counsel; Lexis AI-assisted research features referenced)", "deployer": "Roger Roots (counsel for Joseph Maldonado / Joe Exotic)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Roger Roots, representing the imprisoned Joseph Maldonado (known as \"Joe Exotic\" of Tiger King), submitted filings containing fabricated and inaccurate citations. Roots' paralegal admitted in a sworn declaration that earlier drafts included \"AI-assisted research features\" from Lexis tools. Roots himself pledged to no longer rely on AI-assisted research summaries. The court imposed a $1,500 Rule 11 sanction and dismissed the case for lack of standing.", "canonical_source_conflicted": "Non-existent or misrepresented federal civil rights and animal-law authorities; Fed. R. Civ. P. 11", "mitigation_gap": "Counsel used AI-assisted research tools without adequate verification; admitted through pledged remedial measures and paralegal declaration.", "reliance_or_harm": "$1,500 Rule 11 sanction; case dismissed for lack of Article III standing.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Maldonado v. Professional Animal Retirement Center (PARC), No. 1:25-cv-00454-HAB-ALT (N.D. Ind.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Indiana", "filing_date": "2026-04-01", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.innd.124045/gov.uscourts.innd.124045.26.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Charlotin tracker date (Feb 27) is the Show Cause Order; actual sanctions order is April 1, 2026. Case involves Joseph Maldonado-Passage (Joe Exotic), federally incarcerated.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; lexis-ai-features; dismissed; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-051", "error_title": "Bankruptcy court orders generative-AI CLE and $29,877 in fees over unverified Westlaw Precision citations", "ai_system_name": "Westlaw Precision (named; admitted)", "deployer": "Gregory W. Mitchell (counsel for the Petitioning Creditors)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Gregory W. Mitchell filed motions to quash containing three non-existent direct quotations about Bankruptcy Rule 2004 and citations to non-existent cases. At the show-cause hearing he acknowledged using AI tools available through Westlaw, and his written response admitted his firm has the \"Precision\" version of Westlaw, which the court described as \"a generative AI tool offered by Westlaw.\" Mitchell attributed the text to an unidentifiable prior template whose origin he could not recall, and admitted he had verified none of the citations. The court held causation immaterial: the violation was submitting unverified citations however they were produced.", "canonical_source_conflicted": "Non-existent bankruptcy authorities and fabricated quotations concerning Fed. R. Bankr. P. 2004; Fed. R. Civ. P. 11(b) via Bankr. R. 9011", "mitigation_gap": "Counsel reused an unattributed template and submitted generative-AI-assisted research without verifying a single citation; the court rejected his argument that legal-specific AI tools warrant more lenient treatment than general-purpose ones under its General Orders.", "reliance_or_harm": "Rule 11(b) violation found; six hours of State Bar of Texas continuing legal education on generative AI in the courts ordered, with certificate of compliance due 2026-08-31; civil contempt of the April 30, 2026 turnover order; $29,877.00 in the Chapter 7 Trustee’s attorneys’ fees and costs imposed jointly and severally against counsel and the Petitioning Creditors under the court’s inherent power and 11 U.S.C. § 105(a).", "public_matter_type": "sanctions order (Rule 11 / inherent power) and civil contempt finding", "public_matter_name": "In re Rosslyn2016, LLC, Texas Esencia 2019 LLC and Timbers2020 LLC, No. 25-34507 (Bankr. S.D. Tex., Houston Div.)", "filing_status": "sanctioned", "jurisdiction": "U.S. Bankruptcy Court, Southern District of Texas", "filing_date": "2026-07-14", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.txsb.485264/gov.uscourts.txsb.485264.312.0.pdf", "secondary_source_links": "https://courtlistener.com/docket/71022557/rosslyn2016-llc-and-texas-esencia-2019-llc/", "confidence_score": "high", "notes_on_resolution": "Memorandum Opinion (ECF 312) by Chief Judge Eduardo V. Rodriguez; a conforming separate order was entered simultaneously and is a second primary document. The court did not find that Westlaw Precision produced the fabrications and expressly held causation immaterial, so the AI-establishment basis is counsel’s admission of use, not a judicial finding of tool output. The fee award is compensatory and joint and several with the Petitioning Creditors, not a penalty against counsel alone; the Fifth Circuit bars fee awards for court-initiated Rule 11(c)(1)(B) sanctions, which is why it rests on inherent power and § 105(a). Fee total verified by itemized arithmetic.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; legal-ai-vendor-tool; westlaw-precision; generative-ai-cle; civil-contempt; bankruptcy; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-052", "error_title": "California Court of Appeal sanctions counsel $1,500 in published opinion for AI-supplied fabricated quotations", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Carlton Floyd (counsel for plaintiff and appellant Daniel Del Biaggio)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Carlton Floyd filed an opening brief on appeal containing fabricated quotations attributed to real cases. His declaration admitted his office uses generative AI, and the court found he had used generative AI to add material to the brief during recesses in another trial without ever intending to review the added case law himself or have any other lawyer review it. The court referred to \"the AI-supplied quotations.\" No product is named anywhere in the opinion. Floyd later sent a letter identifying and withdrawing the misstatements.", "canonical_source_conflicted": "Fabricated quotations attributed to genuine California authorities; California Rules of Court rule 8.276", "mitigation_gap": "Counsel delegated substantive brief drafting to a generative-AI tool with no verification step planned at any point, by his own account.", "reliance_or_harm": "$1,500 in appellate sanctions payable to the clerk of the Court of Appeal within 30 days of remittitur; opinion certified for publication; mandatory referral to the State Bar under Cal. Bus. & Prof. Code § 6086.7(a)(3); counsel ordered to serve the opinion on his client and certify service.", "public_matter_type": "sanctions order (appellate, published opinion)", "public_matter_name": "Del Biaggio v. Bansen, No. A174647 (Cal. Ct. App., 1st Dist., Div. 4) (Humboldt County Super. Ct. No. CV1901078)", "filing_status": "sanctioned", "jurisdiction": "California Court of Appeal, First Appellate District", "filing_date": "2026-07-10", "error_date": "2026", "public_record_link": "https://courts.ca.gov/opinions/documents/A174647.PDF", "secondary_source_links": "https://courts.ca.gov/opinions", "confidence_score": "high", "notes_on_resolution": "Certified for publication, making it precedential. In the same opinion the trial court’s separate § 1008(d) sanctions award against counsel was vacated and the respondents’ own frivolous-appeal sanctions motion was denied. Counsel’s declaration hedged that the errors \"were not willful or AI-driven without review,\" and footnote 6 reserves on whether every citation error is attributable to AI, but that reservation covers only uncatalogued additional errors, not the fabricated quotations. Charlotin lists the court as the 4th District; it is the First Appellate District, Division Four, and the tracker PDF filename misspells the caption as \"Del Baggio.\"", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; published-opinion; state-bar-referral; appellate-sanctions; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-053", "error_title": "Eleventh Circuit refers counsel for discipline after eight hallucinated citations were replaced by eight more", "ai_system_name": "Unnamed generative AI (court finding)", "deployer": "Anthony F. Sabatini (counsel for plaintiffs and appellants)", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Counsel filed an opening brief in the Eleventh Circuit containing eight hallucinated cases, one of which was falsely attributed to the Eleventh Circuit itself. He then filed an untimely proposed reply brief purporting to withdraw eight cases, none of which matched the original eight, and all eight substitutes were also hallucinated. The court made its own finding of AI use in the published opinion, stating that counsel had outsourced his legal work to an AI algorithm; no product is named and counsel made no admission, conceding only that his citations were erroneous or unverifiable. The conduct continued after a warning.", "canonical_source_conflicted": "Sixteen non-existent judicial decisions, including one falsely attributed to the Eleventh Circuit; counsel’s duties of competence and candor to the court", "mitigation_gap": "Counsel delegated brief-writing to a generative-AI tool without verification, then attempted to cure the defect using the same unverified process, producing a second complete set of fabrications.", "reliance_or_harm": "Dismissal of the third amended complaint affirmed; published finding that counsel violated his ethical duties to his clients and the court; referral to the Eleventh Circuit’s Committee on Lawyer Qualifications and Conduct by separate order of the Chief Judge under 11th Cir. R., Addendum 8, r. 2(C).", "public_matter_type": "published appellate opinion with disciplinary referral", "public_matter_name": "Akerlund v. Atlas Air, Inc., Flight Services International, LLC, No. 24-11033 (11th Cir.)", "filing_status": "referred for discipline", "jurisdiction": "U.S. Court of Appeals for the Eleventh Circuit", "filing_date": "2026-07-10", "error_date": "2026", "public_record_link": "https://media.ca11.uscourts.gov/opinions/pub/files/202411033.pdf", "secondary_source_links": "https://govinfo.gov/content/pkg/USCOURTS-ca11-24-11033/pdf/USCOURTS-ca11-24-11033-0.pdf", "confidence_score": "high", "notes_on_resolution": "Marked FOR PUBLICATION; appeal from S.D. Fla. No. 1:22-cv-23519-KMM; panel Rosenbaum, Grant, and Kidd, opinion by Grant. Gate 1 rests on the court’s own finding, not an admission: \"By outsourcing his legal work to an AI algorithm, Sabatini violated his ethical duties to both his clients and this Court.\" Charlotin tags the tool \"Implied,\" which understates the record. No monetary sanction. The referral itself issues by a separate order through the Chief Judge, which is a distinct primary document not yet obtained. Tracker PDF filename misspells the defendant as \"Altas Air\" and omits co-defendant Flight Services International, LLC.", "tags": "hallucinated-citations; attorney-discipline; court-found-ai; unnamed-ai; published-opinion; disciplinary-referral; repeat-conduct; appellate; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-054", "error_title": "Georgia Supreme Court sanctions prosecutor whose AI-drafted proposed order put fake citations into a judge’s ruling", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Assistant District Attorney Deborah L. Leslie, Clayton County District Attorney’s Office", "domain": "criminal justice", "error_type": "hallucinated authority", "error_description": "An assistant district attorney used artificial-intelligence software to draft the State’s briefs opposing a motion for new trial, a proposed order for the trial court, and the State’s appellate brief, without independently verifying the citations it produced. The trial court signed the proposed order largely as submitted, so the fabricated authority entered a judge’s signed ruling. The Supreme Court of Georgia identified at least nine defective citations in that order, twelve further cases the prosecutor herself later identified in her trial-court briefing, and nine authorities withdrawn from the State’s appellate brief.", "canonical_source_conflicted": "Non-existent and misattributed Georgia authorities, including a cited decision the Court found appears not to exist at all; Ga. Sup. Ct. R. 7; OCGA 15-1-3(4)", "mitigation_gap": "The prosecutor did not verify AI-generated citations before filing, and the reviewing court signed a proposed order without independent checking — two verification failures compounding, with the second laundering the first into a judicial act.", "reliance_or_harm": "Prosecutor and the Clayton County district attorney’s office admonished; the prosecutor’s privilege to practise before the Supreme Court of Georgia suspended for six months, reinstatement conditioned on 12 hours of continuing legal education beyond normal requirements covering ethics, brief writing, and the proper use of artificial-intelligence software; the trial court’s order denying a new trial vacated and the case remanded for a new order free of fabricated citations.", "public_matter_type": "sanctions order (appellate, published opinion)", "public_matter_name": "Hannah Renea Payne v. The State, No. S26A0459 (Ga.), 324 Ga. 305", "filing_status": "sanctioned", "jurisdiction": "Supreme Court of Georgia", "filing_date": "2026-05-05", "error_date": "2026", "public_record_link": "https://www.gasupreme.us/wp-content/uploads/2026/05/s26a0459.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "The fabricated citations sat in the trial court’s own signed order, but the opinion states that order was largely prepared by the prosecutor as a proposed order, and the AI use is entirely prosecutor-side; the trial judge was neither sanctioned nor found to have used AI. Recorded here as prosecutor-side conduct that contaminated an adjudicative document, not as adjudicator-side AI use. No product named — the opinion says only “artificial intelligence software”. Footnote 4 preserves separate State Bar and Judicial Qualifications Commission proceedings, so a later disciplinary phase is possible. Two justices concurred in part and dissented in part, objecting only to admonishing the district attorney’s office. Official caption spells the appellant “Hannah Renea Payne”. URL note: the bare domain gasupreme.us fails TLS with a certificate hostname mismatch, so this link must retain the www host — see WWW_REQUIRED_HOSTS in scripts/url-policy.mjs.", "tags": "hallucinated-citations; prosecutor; admitted-ai; unnamed-ai; proposed-order-contamination; order-vacated; suspension; criminal-justice", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-055", "error_title": "N.D. Mississippi revokes two attorneys’ pro hac vice admission and fines four lawyers $8,000 over AI-fabricated citations on both sides", "ai_system_name": "\"First Drafts\" (named; admitted) and a second, expressly unnamed in-house legal-research tool", "deployer": "Kathleen M. Wilson (Wilson Law Firm LLC) and Kathryn Y. Williams (Daniel, Williams & Associates PLLC), with resident counsel Shauncey Hunter Ridgeway and Mark C. McClinton", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Counsel for both the plaintiff and the defendant city filed briefs containing citations to non-existent cases. Plaintiff’s counsel admitted using “First Drafts”, an AI-powered legal drafting program, without verifying its output, and admitted using it in filings in other cases. Defence counsel admitted relying on a separate in-house AI research tool, which the order expressly calls “this unnamed software”, also without verification. Both resident attorneys let filings go out under their electronic signatures without review.", "canonical_source_conflicted": "Citations to non-existent decisions, including a fabricated citation attributed to the same district; Fed. R. Civ. P. 11; Local Rule 83.1(c) and (d)(8)", "mitigation_gap": "Two independent AI research tools were used with no verification step, and the local-counsel review layer that exists precisely to catch this failed on both sides of the case simultaneously.", "reliance_or_harm": "Pro hac vice admission revoked for both lead attorneys, each additionally barred from entering an appearance in any case before the district for two years; both resident attorneys disqualified from further participation; fines totalling $8,000 payable to the court registry ($2,500, $3,500, $1,000, $1,000); AI-ethics CLE ordered for the attorney who had not previously attended one; the order transmitted to the Mississippi, Louisiana, and Texas bars, and to the judges presiding over two other matters involving the same counsel; all four attorneys terminated as counsel of record, leaving both litigants unrepresented under a deadline to obtain new counsel or face dismissal.", "public_matter_type": "sanctions order (Rule 11 / inherent power)", "public_matter_name": "Withers v. City of Aberdeen, No. 1:24-CV-218-SA-RP (N.D. Miss.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Mississippi", "filing_date": "2026-06-08", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.msnd.50181/gov.uscourts.msnd.50181.123.0_1.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Sanctions Order, Doc. 123, by Senior U.S. District Judge Sharion Aycock. The $8,000 is an aggregate of four separate per-attorney fines, not a single award. Bad faith was found only as to the two lead attorneys; the resident attorneys were found negligent and careless but not in bad faith. Do not attribute the defence tool to “First Drafts” — the order treats them as two distinct tools and names no vendor for either. Related but distinct corpus record: AIEL-2026-031 (In re Wise, Bankr. W.D. La.) involves the same plaintiff-side attorney; this order discusses that matter as later repeat conduct, and the Withers conduct predates it.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; first-drafts; pro-hac-vice-revoked; local-counsel-liability; multi-bar-referral; both-sides; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-056", "error_title": "S.D.N.Y. publicly admonishes counsel for hallucinated citations after LexisNexis denied he had the AI tool he blamed", "ai_system_name": "Protégé / Lexis+ AI (named by counsel; vendor denied he had access)", "deployer": "Tyrone A. Blackburn (T.A. Blackburn Law, PLLC), defence counsel and himself a named defendant", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Counsel’s brief supporting a motion to dismiss cited several non-existent authorities and misstated the law. He did not dispute this and explained that the miscitations came from using the LexisNexis AI research platform Protégé. LexisNexis then provided the court a letter refuting that he held any subscription to Lexis+ AI or Protégé, so the court recorded concern both about the hallucinated authority and about the attorney’s misrepresentation of where his research came from.", "canonical_source_conflicted": "Non-existent judicial authority and misstatements of law in briefing on a Rule 12(b)(6) motion", "mitigation_gap": "Counsel filed AI-derived research without verification and then attributed it to a product he did not have, so the account of the failure was itself unreliable.", "reliance_or_harm": "Public admonition recorded in a published opinion and order, expressly intended to deter future misconduct, with a warning that further misconduct may lead to sanctions or a disciplinary referral; the court declined to strike the submissions or award fees.", "public_matter_type": "opinion and order with public admonition", "public_matter_name": "Joseph Cartagena v. Terrance Dixon, Tyrone Blackburn, and T.A. Blackburn Law, PLLC, No. 1:25-cv-03552 (JLR) (S.D.N.Y.)", "filing_status": "admonished", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2026-03-10", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.641455/gov.uscourts.nysd.641455.161.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Opinion and Order, Dkt. 161, by Judge Jennifer L. Rochon; the AI discussion is a threshold section preceding the merits. This is the first AI-related judicial finding in this matter and the one that carries gate-1 weight: two later orders (Dkt. 206 of 2026-06-23 and an order of 2026-07-10) reference AI only as allegation, denial, or recitation of sanctions in other matters, and the Grievance Committee referral that followed rests on deposition conduct rather than AI use. Record the vendor attribution as claimed-and-denied, not as established. Related matters involving the same attorney, none yet in the corpus: Jakes v. Youngblood (W.D. Pa.), Facey v. Fisher (N.Y. Sup. Ct.), and Flycatcher Corp. v. Affable Ave. LLC (S.D.N.Y.).", "tags": "hallucinated-citations; attorney-discipline; named-ai; legal-ai-vendor-tool; vendor-denied-access; misrepresentation-to-court; public-admonition; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-057", "error_title": "GAO warns bid-protest counsel who filed Westlaw CoCounsel citations to non-existent decisions", "ai_system_name": "Westlaw CoCounsel (named; admitted)", "deployer": "Counsel for the protester CVTEK, LLC (McDonald Hopkins LLC)", "domain": "government procurement", "error_type": "hallucinated authority", "error_description": "In a supplemental bid protest, counsel filed citations that the Government Accountability Office could not locate at all, alongside citations whose real decisions did not state the propositions claimed. GAO issued a formal Notice of Required Protester Response compelling counsel to justify the citations or produce the decisions. One of the protester’s attorneys confirmed he had used Westlaw’s AI tool CoCounsel in preparing the supplemental protest, the attorneys acknowledged some of its citations were erroneous, and the attorney admittedly failed to correct all of them before filing.", "canonical_source_conflicted": "Non-existent GAO and related procurement decisions, plus real decisions cited for propositions they do not state; 4 C.F.R. Part 21", "mitigation_gap": "AI-generated citations were filed without verification, and errors identified before filing were still not fully corrected in the version submitted.", "reliance_or_harm": "GAO made findings on the record in a dedicated section of a published decision and expressly advised the protester that future filings citing non-existent authority may result in sanctions; it declined to impose sanctions because it denied the protest on independent grounds. Counsel adopted new manual citation-verification protocols.", "public_matter_type": "administrative adjudication with formal advisement", "public_matter_name": "Matter of: CVTEK, LLC, B-423943; B-423943.2 (U.S. Government Accountability Office)", "filing_status": "warned", "jurisdiction": "U.S. Government Accountability Office", "filing_date": "2026-02-12", "error_date": "2026", "public_record_link": "https://gao.gov/assets/890/884082.pdf", "secondary_source_links": "https://gao.gov/products/b-423943,b-423943.2", "confidence_score": "high", "notes_on_resolution": "Broadens the corpus beyond court sanctions into federal administrative procurement adjudication. GAO issues protest decisions and does not impose Rule 11 sanctions, so the outcome is an on-the-record advisement plus a compelled-response order, not a sanction; the fabricated citations did not affect the result, as both protest grounds were denied on independent merits grounds. Underlying procurement: DHS/USCIS RFP No. 70SBUR25R00000009. The secondary landing page must keep the comma-joined file numbers — the shortened form is a hard 404. Tooling note: gao.gov returns 403 to plain scripted requests behind Akamai for both bare and www hosts, so an automated link check failing on this URL is a known false negative; the asset was confirmed by hash against an independently obtained copy. Three further GAO matters cited in this decision are queued leads: KE Systems Services (B-423881), Oready, LLC (B-423524.2), and Raven Investigations & Security Consulting (B-423447) — note the Oready number differs from the one recorded in AIEL-CAND-004.", "tags": "hallucinated-citations; named-ai; legal-ai-vendor-tool; westlaw-cocounsel; administrative-adjudication; bid-protest; warning; government-procurement", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-24", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-058", "error_title": "W.D. Pennsylvania fines counsel $5,000 and revokes pro hac vice status for AI-fabricated authorities", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Tyrone Blackburn, counsel for defendant Duane Youngblood", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Tyrone Blackburn admitted that his use of generative AI to draft briefs caused inaccurate citations and quotations, including inaccurate quotations from the court's own earlier opinion. The court rejected his explanation that the defects were an innocent mistake caused by technological ignorance, emphasized that he doubled down after opposing counsel identified the errors, and found his conduct sanctionable under Federal Rule of Civil Procedure 11.", "canonical_source_conflicted": "Accurate judicial authorities and quotations; Fed. R. Civ. P. 11", "mitigation_gap": "Counsel used generative AI to draft briefing without verifying its citations and quotations, then repeated the defective material after the problem was raised.", "reliance_or_harm": "The court imposed a $5,000 fee sanction payable in monthly installments of at least $500, revoked Blackburn's pro hac vice status, and required any future appearance or pro hac vice motion he files in the Western District of Pennsylvania to include the sanctions order.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Jakes v. Youngblood, No. 2:24-cv-01608-WSS (W.D. Pa.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Western District of Pennsylvania", "filing_date": "2025-10-06", "error_date": "2025", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.pawd.314851/gov.uscourts.pawd.314851.71.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Memorandum Order, ECF No. 71, by Judge William S. Stickman IV. The native image-only order was verified through RECAP against its caption, docket, filing date, entry number, and PACER case ID, then OCR-reviewed across all six pages. Corrects the review draft's mistaken first name: the sanctioned attorney is Tyrone Blackburn, not Arthur Aidala Blackburn.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; rule-11; pro-hac-vice-revoked; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-25", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-059", "error_title": "N.D. Mississippi disqualifies three lawyers and orders firm-wide audit after admitted Grok misuse", "ai_system_name": "Grok (named; admitted)", "deployer": "Jane Ashley Watson, with supervising attorneys Nick Norris and Louis H. Watson Jr. and Watson & Norris, PLLC", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Jane Ashley Watson admitted using Grok, an external AI tool, for legal research and drafting without verifying its output. Her summary-judgment memorandum contained one nonexistent case and three misrepresented holdings. The court found Rule 11 violations by Watson and reviewing signatory Nick Norris, held Watson & Norris, PLLC jointly responsible, and sanctioned supervisory attorney Louis H. Watson Jr. after finding that the firm failed to contain a broader pattern of tainted filings despite prior notice and an internal policy restricting external AI use.", "canonical_source_conflicted": "One nonexistent decision, three materially misrepresented holdings, and counsel's duties under Fed. R. Civ. P. 11", "mitigation_gap": "The drafting attorney used unverified Grok output despite prior warning and a firm policy against external AI tools; supervising and signing attorneys did not verify the authorities, and the firm's later audit initially failed to identify the full scope of affected filings.", "reliance_or_harm": "The court disqualified all three attorneys from the case, barred Jane Watson from new appearances before the judge for two years, required notice to every judge in their pending matters, referred the order to the Mississippi Bar, ordered a firm-wide audit, and stayed the client's case for 60 days after the citation failures delayed summary judgment and caused cancellation of the trial setting.", "public_matter_type": "sanctions order (Rule 11 / inherent power)", "public_matter_name": "Billups v. Louisville Municipal School District, No. 1:24-cv-00074-SA-RP (N.D. Miss.)", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Mississippi", "filing_date": "2025-12-19", "error_date": "2025", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.msnd.49169/gov.uscourts.msnd.49169.79.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Sanctions Order, ECF No. 79, by Senior Judge Sharion Aycock. The native order was verified through RECAP against its caption, docket, filing date, entry number, and PACER case ID. The order names Grok as Watson's external AI tool and separately describes the firm's permitted Smokeball tool, Archie; do not attribute the defective filing to Archie. The court found a monetary fine would not provide meaningful deterrence and imposed structural and practice restrictions instead.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; grok; rule-11; disqualification; firm-audit; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-07-25", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-060", "error_title": "Illinois Appellate Court fines attorney $15,000 for ten AI-hallucinated citations and quotations", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Mason Cole, of Cole Sadkin, LLC, counsel for petitioner Kimberly Scott", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Mason Cole filed appellate briefs on behalf of a workplace-discrimination petitioner containing nonexistent case citations and fabricated statutory language. In response to the court's show-cause order he stated that he uses a premier corporate subscription of ChatGPT to synthesize complex matters, including this case, and that he had cross-referenced citations with LexisNexis but overlooked that several did not exist or did not support the propositions cited. The court identified ten distinct false citations or misstatements of law.", "canonical_source_conflicted": "Nonexistent Illinois appellate authority, fabricated statutory language including a supposed 'full investigation' requirement under the Illinois Human Rights Act, and counsel's duties under Illinois Supreme Court Rules 341(h)(5) and 375(b)", "mitigation_gap": "Counsel relied on ChatGPT output for appellate research and represented that he cross-checked citations against LexisNexis, but that verification did not catch citations that were nonexistent or that failed to support the cited proposition; he attributed the failure to inexperience in appellate practice.", "reliance_or_harm": "The court affirmed the underlying dismissal, fined counsel $15,000 payable to the clerk of the Appellate Court within 30 days at a rate of $1,500 per false citation or misstatement, and directed the clerk to send a copy of the opinion to the Illinois Attorney Registration and Disciplinary Commission. The court recorded that the misconduct forced it to verify the fabricated citations, research nationwide sanctions law, and order and review counsel's response.", "public_matter_type": "sanctions order (Illinois Supreme Court Rule 375(b))", "public_matter_name": "Scott v. Illinois Human Rights Commission, 2026 IL App (1st) 251462, No. 1-25-1462", "filing_status": "sanctioned", "jurisdiction": "Appellate Court of Illinois, First District", "filing_date": "2026-07-28", "error_date": "2026", "public_record_link": "https://illinoiscourts.gov/resources/23d7df84-ed51-48df-8477-3a6872db40d5/file", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Published opinion filed 28 July 2026, Second Division, Presiding Justice Van Tine. Verified against the Illinois courts' own resource copy; caption, docket number, neutral citation and filing date all match. The $1,500-per-item rate is the court's own stated calculation and is higher than the $1,000-per-fabrication rate it surveyed from other Illinois decisions.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; chatgpt; monetary-sanction; ardc-referral; appellate; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-16", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-061", "error_title": "D. Connecticut imposes $3,500 Rule 11 sanction over Open Law, Claude and ChatGPT hallucinations", "ai_system_name": "Open Law, Claude and ChatGPT (named; admitted)", "deployer": "Hilary Miller, defense counsel for Tacobarn Newtown LLC and Tacobarn Greenwich LLC", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Defense counsel Hilary Miller filed a motion to dismiss and a motion to exclude extrinsic evidence containing fabricated citations and fabricated quotations from real cases. At the show-cause hearing he stated that this was the first and only matter in which he used artificial intelligence, that he used a paid tool called Open Law for argument and case suggestions, and that he also used Claude and ChatGPT for additional suggestions. He Shepardized every case through Lexis but did not verify that pin cites referred to the correct page or that quotations were accurate.", "canonical_source_conflicted": "Fabricated quotations attributed to Ashcroft v. Iqbal and other real decisions, together with nonexistent supporting authority, against counsel's certification duties under Fed. R. Civ. P. 11(b)(2)", "mitigation_gap": "Counsel's verification step confirmed that the cited cases existed and remained good law but did not check whether the quoted language or pin cites were real, which is precisely the failure mode generative AI produces; the court found the inadequate inquiry weighed in favour of sanctions.", "reliance_or_harm": "The court imposed a $3,500 monetary sanction and referred the matter to the Grievance Committee under its local rules. The court set out a three-category definition of AI hallucination in legal research and warned that hallucinated or severely misstated law will often result in sanctions regardless of whether AI tools were used.", "public_matter_type": "sanctions order (Rule 11, sua sponte)", "public_matter_name": "Barteca Holdings LLC v. Tacobarn Newtown LLC, No. 3:26-cv-00250-VDO (D. Conn.), ECF No. 43", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Connecticut", "filing_date": "2026-08-04", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.ctd.169303/gov.uscourts.ctd.169303.43.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Order Imposing Sanctions by Judge Vernon D. Oliver. The native order was resolved through the CourtListener RECAP API using the PACER case id (ctd.169303) and the filepath_local value for entry 43, then verified against its caption, docket number, entry number and 08/04/26 filing stamp. Open Law is a paid legal-research tool distinct from the general-purpose assistants; the order names all three.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; claude; chatgpt; open-law; rule-11; monetary-sanction; grievance-referral; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-16", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-062", "error_title": "W.D. Washington sanctions counsel $3,000 for ChatGPT and Claude citations after repeated warnings", "ai_system_name": "ChatGPT and Claude (named; admitted)", "deployer": "Jocelyn Stewart, counsel for plaintiff Joann LeDoux", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Between November and December 2025 plaintiff's counsel Jocelyn Stewart filed multiple submissions containing dozens of inaccurate factual and legal citations. She later averred that she used Claude and ChatGPT to draft the submissions, that the programs generated the false citations, and that she did not review their outputs because she believed the tools operated in a closed system limited to documents she supplied. She continued to submit a hallucinated citation in December, after opposing counsel had first raised AI concerns at oral argument.", "canonical_source_conflicted": "Dozens of nonexistent or materially altered case citations and quotations whose meaning had been changed, against counsel's certification duties under Fed. R. Civ. P. 11(b)", "mitigation_gap": "Counsel adopted a mistaken belief that a closed-system retrieval setup could not hallucinate and therefore performed no verification of AI output, despite having previously heard of an attorney sanctioned for the same conduct and despite numerous Ninth Circuit courts having issued or threatened comparable sanctions before her filings.", "reliance_or_harm": "The court found the conduct akin to contempt, imposed a $3,000 personal sanction on counsel payable within 21 days, and required her to provide a copy of the order to her client and certify that she had done so. Counsel had earlier been required to withdraw or amend the affected motions and ceased using AI to generate quotations or citations for filings. On 18 August 2026 the court granted Defendants' Daubert motions, excluding both of Plaintiff's remaining experts. It found that multiple hallucinated citations to academic articles in Dr. Gabriel Holguin's report irreparably undermined his credibility and that Plaintiff had not met her burden to show the opinion was based on sufficient facts or data, citing among other things the AI-generated list of materials supposedly relied upon and Dr. Holguin's lack of diligence. The court then granted summary judgment and dismissed Plaintiff's remaining claims with prejudice.", "public_matter_type": "sanctions order (Rule 11, sua sponte)", "public_matter_name": "LeDoux v. Outliers, Inc., No. 3:24-cv-05808-TMC (W.D. Wash.), ECF No. 265", "filing_status": "sanctioned; case dismissed with prejudice", "jurisdiction": "U.S. District Court, Western District of Washington", "filing_date": "2026-07-24", "error_date": "2025", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.wawd.339711/gov.uscourts.wawd.339711.265.0.pdf", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.wawd.339711/gov.uscourts.wawd.339711.269.0.pdf", "confidence_score": "high", "notes_on_resolution": "Order on Sanctions by Judge Tiffany M. Cartwright, discharging the 4 February 2026 show-cause order at ECF No. 209. The native order was resolved through the CourtListener RECAP API using the PACER case id (wawd.339711) and the filepath_local value for entry 265, then verified against caption, docket number, entry number and 07/24/26 filing stamp. The closed-system belief is recorded by the court as counsel's explanation, not as a finding that such systems cannot hallucinate. Updated 19 August 2026 for the Order Granting Defendants' Motion for Summary Judgment at ECF No. 269, filed 18 August 2026, resolved through the RECAP path for entry 269 and byte-identical to the tracker's copy at 304,135 bytes. That order extends the incident from counsel-side to expert-side AI use: the fabricated citations reached an expert report, the expert was excluded, and the case ended. It is the corpus's first record in which AI-hallucinated authority is the stated basis for excluding an expert witness.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; chatgpt; claude; rule-11; monetary-sanction; closed-system-misconception; legal-services; expert-witness; daubert-exclusion; dismissed-with-prejudice", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-19", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-063", "error_title": "Connecticut Supreme Court sanctions attorney and firm in first generative-AI hallucination case", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Ian G. Gottlieb and GLG Law, LLC", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Attorney Ian G. Gottlieb and his firm filed applications for certification to file public interest appeals under General Statutes section 52-265a in two related landlord-tenant matters. The applications contained hallucinated citations produced by generative AI used during the editing and review process, and after the Chief Justice granted the applications the briefs filed in each case also contained hallucinated citations. At a show-cause hearing counsel candidly admitted that filing approximately seven erroneous and unverified citations as a result of generative-AI use violated at a minimum rule 1.1 of the Rules of Professional Conduct.", "canonical_source_conflicted": "Approximately seven fabricated or inaccurately generated citations, against rule 1.1 of the Connecticut Rules of Professional Conduct and counsel's duty of candour to the court", "mitigation_gap": "Generative AI was introduced at the editing and review stage rather than at drafting, so the output bypassed the verification counsel had applied earlier; counsel had initially used LexisNexis for research and the defective citations entered through the later AI-assisted pass.", "reliance_or_harm": "The court ordered Attorney Gottlieb to complete six hours of continuing legal education in ethics and law office management beyond his annual requirement, three of which must concern generative AI, and ordered Gottlieb and the law firm each to donate $1,000 to the CT Bar Institute, Inc. Both must file a compliance report with the Appellate Clerk and send a courtesy copy to the Statewide Grievance Committee.", "public_matter_type": "sanctions order (inherent supervisory authority)", "public_matter_name": "TOV Realty, LLC v. Suarez (SC 21183) and Kosel Equity, LLC v. MacGregor (SC 21184), 355 Conn. 902 (2026)", "filing_status": "sanctioned", "jurisdiction": "Connecticut Supreme Court", "filing_date": "2026-07-31", "error_date": "2026", "public_record_link": "https://jud.ct.gov/external/supapp/Cases/AROcr/CR355/ORD355.3.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Per curiam order of 31 July 2026, the Connecticut Supreme Court's first generative-AI hallucination sanction, consolidating two docket numbers. Verified against the Judicial Branch's own advance-release copy. The order names no AI product, so the record uses the unnamed-but-admitted qualifier. The monetary component is a charitable donation of $1,000 each by counsel and the firm, not a fine, and the third-party hallucination tracker recorded only a single $1,000 figure and omitted the firm's parallel obligation.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; state-supreme-court; cle-requirement; charitable-donation; grievance-committee; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-16", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-064", "error_title": "S.D. Mississippi admonishes counsel over fictitious citation traced to an unconfirmed Google AI preview", "ai_system_name": "Google AI preview (raised in the record; attribution not determined)", "deployer": "Aaron Randall Rice, counsel for plaintiff Johnny Fuselier", "domain": "legal services", "error_type": "fabricated authority (AI attribution undetermined)", "error_description": "In a 5 February 2026 response opposing summary judgment, counsel cited \"Rodriguez v. It's Just Lunch, Int'l, 2013 WL 12173926 (5th Cir. 2013),\" a case that does not exist: the reporter number belongs to Vasudevan Software, Inc. v. Microstrategy, Inc. and the caption to an unrelated S.D.N.Y. matter. The brief also attributed to Minnesota Voters Alliance v. Mansky language the Supreme Court did not use. Ordered to disclose whether and how he had used artificial intelligence, counsel explained that he had replaced his Westlaw subscription with Fastcase, was unfamiliar with it, and turned to Google web search to locate authority, intending to verify citations in Fastcase afterwards. He obtained the Rodriguez citation through those searches but was unable to confirm its source, stating he was \"not certain it came from the AI preview, simply because [he is] skeptical of AI responses and do[es] not use interactive AI tools for legal research,\" while acknowledging that because the citation refers to a nonexistent opinion \"it may have appeared there.\" The court recounted this account without rejecting it and did not determine the citation's origin.", "canonical_source_conflicted": "A nonexistent Fifth Circuit decision and a quotation not present in Minnesota Voters Alliance v. Mansky, against counsel's certification duties under Fed. R. Civ. P. 11", "mitigation_gap": "Counsel changed research platforms, was unfamiliar with the replacement, and used general web search as an interim substitute for a legal database. He planned to verify citations in Fastcase but, rushing to meet the filing deadline, did not verify this one. The record shows no verification step between retrieving a citation from web search results and filing it.", "reliance_or_harm": "The fictitious citation and misattributed quotation reached a summary-judgment response brief in a First Amendment case. On review of counsel's show-cause response the court found an admonishment sufficient and imposed no sanction, citing the limited nature of the violations, counsel's ownership of the mistakes, and changes he had implemented to prevent recurrence. Counsel was admonished and warned that future conduct of this kind will result in sanctions.", "public_matter_type": "show-cause order and admonishment (Rule 11)", "public_matter_name": "Fuselier v. Riscassi, No. 1:25-cv-268-HSO-BWR (S.D. Miss.), ECF No. 28", "filing_status": "admonished; no sanction imposed", "jurisdiction": "U.S. District Court, Southern District of Mississippi", "filing_date": "2026-05-22", "error_date": "2026-02", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.mssd.130336/gov.uscourts.mssd.130336.28.0.pdf", "secondary_source_links": "https://storage.courtlistener.com/recap/gov.uscourts.mssd.130336/gov.uscourts.mssd.130336.25.0.pdf", "confidence_score": "medium", "notes_on_resolution": "Order of Chief District Judge Halil S. Ozerden, 22 May 2026, resolving the 1 May 2026 show-cause order at ECF No. 25. Admitted under the undetermined-attribution rule in INTENT.md: AI is raised in the primary source and remains undisplaced, but no tribunal finding attributes the citation to it. Counsel's own words, quoted by the court, are that he was \"not certain it came from the AI preview\" though \"it may have appeared there.\" The court itself made no attribution finding, observing only that \"[i]n the age of artificial intelligence, all attorneys must remember their obligations under Rule 11.\" The implicated mechanism is passive exposure to an AI-generated summary in general web search results rather than use of a legal-research AI tool, which distinguishes this record from the corpus's tool-attributed sanctions and connects it to AIEL-GLOB-013 on Google AI Overviews. Both orders resolved through the CourtListener RECAP API and verified against caption, docket number, entry number and filing stamp.", "tags": "fabricated-citations; attribution-not-determined; google-ai-overviews; passive-ai-exposure; rule-11; admonishment; no-sanction; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-19", "needs_review": "no" }, { "error_id": "AIEL-2026-065", "error_title": "Dutch disciplinary tribunal reprimands advocate over unverified AI-generated case references in two tenancy matters", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Respondent advocate (name not published), Netherlands", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Acting in two tenancy matters outside her usual employment-law practice, the respondent advocate drafted conclusions of answer using an AI tool and filed them without checking their output. The filings contained incorrect references to partly nonexistent case law, including ECLI citations that proved untraceable on verification. The dean of the local bar obtained the filings and correspondence, questioned her, and lodged a dean's objection under article 46 Advocatenwet. Asked which AI tool she had used in these matters, the advocate answered that she had not been able to determine which tool it was; the last tool she had used was LegalPA, which she had cancelled the previous month. She stated she had not checked whether the decisions the tool referred to existed and had not been alert to the risk that AI tools invent case law.", "canonical_source_conflicted": "Cited Dutch judicial decisions, referenced by ECLI, that proved untraceable or partly nonexistent on verification, against the core values of expertise and integrity in article 10a Advocatenwet", "mitigation_gap": "The advocate accepted instructions in an area of law in which she had neither training nor experience, treated AI output as she would search-engine results, and performed no verification of the cited decisions before filing. She was not alert to the risk that an AI tool would generate nonexistent case law, and she did not appear at the hearings in either matter.", "reliance_or_harm": "The fabricated references reached filed conclusions of answer in two tenancy proceedings in which the advocate also failed to appear, and in one of which she raised no defence to the alternative claim. The tribunal declared the dean's objection unfounded as to the core value of confidentiality and well-founded on the remainder, imposed the measure of reprimand, and ordered payment of EUR 1,250 in costs of the proceedings, comprising EUR 750 to the Netherlands Bar and EUR 500 to the State, payable within four weeks of the decision becoming irrevocable.", "public_matter_type": "disciplinary decision on a dean's objection", "public_matter_name": "Beslissing in de zaak 26-379/DB/LI/D, ECLI:NL:TADRSHE:2026:93 (Raad van Discipline in het ressort 's-Hertogenbosch)", "filing_status": "reprimanded; costs ordered", "jurisdiction": "Raad van Discipline 's-Hertogenbosch", "filing_date": "2026-07-27", "error_date": "2025", "public_record_link": "https://tuchtrecht.overheid.nl/zoeken/resultaat/uitspraak/2026/ECLI_NL_TADRSHE_2026_93", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Decision of 27 July 2026, Raad van Discipline in het ressort 's-Hertogenbosch (mr. E. Loesberg presiding), on the official Dutch government tuchtrecht publication. Promoted from AIEL-GLOB-016 under the jurisdiction rule in INTENT.md: source language does not block admission. Operative language quoted in the original with English renderings. Measure: \"legt aan verweerster de maatregel van berisping\" (imposes on the respondent the measure of reprimand). Costs: \"veroordeelt verweerster tot betaling van de proceskosten van EUR 1.250,- aan de Nederlandse Orde van Advocaten\", split at 7.1 into \"EUR 750,-\" for the Netherlands Bar and \"EUR 500,- kosten van de Staat\". This is a costs order (proceskosten), not a fine; a third-party tracker recorded it as a monetary fine and that characterisation is not adopted. On AI attribution, the tool is NOT named for these matters: asked \"Welke AI-tool heeft u gebruikt in de hiervoor genoemde zaken?\" the advocate answered \"Ik heb niet kunnen achterhalen welke AI-tool ik heb gebruikt\" (I have not been able to determine which AI tool I used), and identified LegalPA only as \"de laatste tool die ik heb gebruikt\" (the last tool I used), cancelled the month before. An earlier draft of this record attributed the matter to LegalPA; that overstated the source and has been corrected. The decision records the defect as \"onjuiste verwijzingen naar (deels niet bestaande) rechtspraak\" (incorrect references to partly nonexistent case law) and ECLI references that \"na controle niet traceerbaar bleken\" (proved untraceable on verification). The confidentiality limb of the objection was rejected because she entered no confidential data into the tool.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; reprimand; costs-order; tenancy-law; netherlands; source-language-dutch; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-08-19", "needs_review": "no" }, { "error_id": "AIEL-2026-066", "error_title": "N.D. Georgia sanctions counsel $8,000 for AI citations after he denied, then admitted, AI use", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "Plaintiff's counsel (identified in the record only as \"Mr. Johnson\"), counsel for plaintiff Erin Booker", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Plaintiff's counsel filed briefs in a slip-and-fall suit against Kroger containing fabricated case citations and misrepresentations of real authority. At a June 26, 2026 hearing, when directly questioned, counsel denied using artificial intelligence. In his later response to the court's order to show cause, he acknowledged that he had \"consulted\" AI tools in drafting the filings.", "canonical_source_conflicted": "Four fake, false, or misleading case authorities cited in filings, and counsel's direct denial of AI use under questioning at a June 26, 2026 hearing", "mitigation_gap": "Counsel did not verify AI-generated citations before filing them, and when first asked about AI use in open court he denied it rather than disclosing it; the false denial, not merely the fabrications, is what the court sanctioned most heavily.", "reliance_or_harm": "The court imposed $1,000 per fake, false, or misleading authority (four total), doubled to $8,000 for counsel's false denial, payable to the court registry's general sanctions fund within 30 days. In lieu of ordering additional AI training, the court instead ordered counsel to file, within 7 days, documentation verifying the ethics-and-technology CLE training he claimed to have already completed, and reserved the question of further sanctions. Defendant's request for attorney's fees was declined as too difficult to apportion.", "public_matter_type": "sanctions order (court's own order to show cause)", "public_matter_name": "Erin Booker v. The Kroger Co., No. 1:26-cv-02006-SDG (N.D. Ga.), ECF No. 66", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Northern District of Georgia", "filing_date": "2026-08-28", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.gand.358237/gov.uscourts.gand.358237.66.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Opinion and Order of Sanctions by Judge Steven D. Grimberg, resolved through the CourtListener RECAP path (PACER case id gand.358237, entry 66). No AI product is named anywhere in the order — \"ChatGPT\" appears only inside the title of a CLE course counsel claimed to have completed (\"ChatGPT, Alexa, and Siri Are Out to Get You!\") — so the tracker's implied product tag does not survive. The tracker's outcome tag (\"Monetary Sanctions; CLE\") is inaccurate: the court expressly declined to order additional CLE and instead required documentation of previously completed training.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; false-denial; monetary-sanction; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-02", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-067", "error_title": "Florida's Second DCA sanctions attorney for 46 AI-hallucinated citations across two briefs", "ai_system_name": "Unnamed generative AI (admitted by counsel; court finding)", "deployer": "Kenneth H. Keefe, The Keefe Law Firm, P.A., South Pasadena, Florida, counsel for appellants", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Counsel for appellants in a mortgage-foreclosure appeal filed an amended initial brief and a reply brief containing forty-six defective citations to case law: thirty-one in the amended initial brief and fifteen in the reply, including one wholly fabricated case and multiple real cases cited for propositions they do not support. Counsel admitted using AI as a research tool without independently verifying the citations it produced. The court found generative AI was not merely his \"supplemental research tool\" but his \"primary or only one.\"", "canonical_source_conflicted": "Forty-six fabricated or misrepresented case citations, including the wholly fictitious Board of Regents v. Wilson, 365 So. 2d 213 (Fla. 1st DCA 1979), against the appellate rules of professional candor", "mitigation_gap": "Counsel used generative AI \"without any apparent guardrails,\" per the court's finding, and did not independently verify any of the AI-generated authorities through reliable legal research sources before filing.", "reliance_or_harm": "The court struck the amended initial and reply briefs, fined counsel $1,500 ($500 for a belated show-cause response plus $1,000 for the AI misuse) payable to the clerk within 14 days, made counsel personally liable for the opposing party's appellate attorney's fees for researching and responding to the defective briefs (amount remanded to the trial court to determine), referred counsel to The Florida Bar, and expressly barred him from charging his clients for either the fine or the fee award. The underlying judgment was affirmed on the merits based on later, corrected briefing.", "public_matter_type": "sanctions order (appellate court's inherent authority)", "public_matter_name": "Capital Standard, LLC v. U.S. Bank National Association, as trustee for Bear Stearns Asset Backed Securities I Trust 2005-AC9, No. 2D2024-1392 (Fla. 2d DCA)", "filing_status": "sanctioned; judgment affirmed", "jurisdiction": "District Court of Appeal of Florida, Second District", "filing_date": "2026-08-21", "error_date": "2026", "public_record_link": "https://flcourts-media.flcourts.gov/content/download/2494161/opinion/Opinion_2024-1392.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Opinion by Rothstein-Youakim, J. (Northcutt and Smith, JJ., concurring), marked subject to revision prior to official publication at the time of verification — recheck for a final So. 3d citation. Counsel admitted using AI as \"a supplemental research tool\" but did not \"fully appreciate that such tools can generate inaccurate or nonexistent citations.\" The opinion itself cites Damien Charlotin's hallucination database at footnote 2 and discusses the June 15, 2026 amendment to Fla. R. Gen. Prac. & Jud. Admin. 2.515(d), which now requires an express representation that cited authorities exist and are accurately cited.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; unnamed-ai; monetary-sanction; fee-shift; bar-referral; briefs-struck; appellate; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-02", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-068", "error_title": "D. Arizona reprimands attorney for ChatGPT-fabricated quotations across four briefs", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Elizabeth Tate, counsel for plaintiff Michael L. Ruiz", "domain": "legal services", "error_type": "hallucinated authority", "error_description": "Plaintiff's counsel used ChatGPT to draft portions of a fee-response brief and other filings, producing a false direct quotation attributed to a Ninth Circuit case (Osazuwa) and additional fabricated quotations across three further briefs, plus an inaccurate AI-generated case summary of a Third Circuit case (Ansell v. Green Acres). This was the terminal order in a three-stage sanctions proceeding that began with a July 8, 2026 order to show cause and a July 28, 2026 order finding the errors attributable to generative-AI misuse.", "canonical_source_conflicted": "Multiple AI-generated false quotations and an inaccurate case summary across four separate briefs, cited to non-existent supporting language in real Ninth and Third Circuit opinions", "mitigation_gap": "Counsel did not verify AI-generated quotations against the actual case text before filing, across multiple briefs and after at least one earlier warning within the same proceeding.", "reliance_or_harm": "The court issued a formal public reprimand of counsel, which she must report in any future context requiring disclosure of past attorney discipline. The court cited prior discipline, a pattern of misconduct, and counsel's substantial experience as aggravating factors. No monetary sanction, CLE requirement, or State Bar referral was ordered in this final order.", "public_matter_type": "sanctions order (court's own show-cause proceeding)", "public_matter_name": "Michael L. Ruiz v. Magellan Financial & Insurance Services, No. CV-23-02090-PHX-DWL (D. Ariz.), Doc. 179", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Arizona", "filing_date": "2026-08-24", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.azd.1348340/gov.uscourts.azd.1348340.179.0.pdf", "secondary_source_links": "", "confidence_score": "high", "notes_on_resolution": "Order by Judge Dominic W. Lanza, resolved through the CourtListener RECAP path (PACER case id azd.1348340, entry 179). This is the terminal order of a three-stage proceeding (Doc. 163 order to show cause, Doc. 172 finding of AI misuse, Doc. 179 final reprimand); no earlier phase of this matter previously existed in the corpus. Note: the hallucination tracker's separate entry for an earlier July 28, 2026 phase of this same matter attributes one of the fabricated quotations to \"Claude AI,\" but this order quotes counsel's own admission that she \"utilized ChatGPT\" for that same brief — treat any future Claude attribution for this matter with caution pending re-verification against Doc. 172.", "tags": "hallucinated-citations; attorney-discipline; admitted-ai; named-ai; chatgpt; reprimand; legal-services", "source_quality": "primary record", "research_status": "included", "last_verified_date": "2026-09-02", "needs_review": "no", "obligation_first_anchors": [ "https://everyailaw.com/obligation-category/human-oversight.json" ] }, { "error_id": "AIEL-2026-069", "error_title": "USPTO publicly reprimands patent attorney over AI-generated citations to a patent intrinsic record", "ai_system_name": "Two unnamed generative AI tools, one for drafting and one for review (neither identified in the order)", "deployer": "Brian E. Mitchell, registered patent attorney (Registration Number 50,612)", "domain": "legal services", "error_type": "fabricated and misattributed citations to the intrinsic record", "error_description": "The final order records that, preparing the plaintiff's portion of a Joint Claim Construction Chart in Magpul Industries Corp. v. Mission First Tactical Group, Inc., No. 2:24-cv-5551 (E.D. Pa.), the respondent \"used one generative artificial intelligence (“AI”) tool to assist with drafting proposed claim constructions of certain claim terms of the patent in question\" and \"then used a second AI tool to review\" that portion of the document. The AI-generated sections cited the patent's intrinsic evidence, including the specification, figures and prosecution history, and at least some of those citations were erroneous: the parties determined that many of the plaintiff's \"citations, quotations, and parentheticals referred to portions of the record that did not exist or were inaccurately attributed.\" The order does not name either AI tool.", "canonical_source_conflicted": "The patent's own intrinsic record, comprising the specification, figures and prosecution history, against citations and quotations that did not exist in it or were inaccurately attributed to it", "mitigation_gap": "The respondent acknowledged that the erroneous citations \"were not sufficiently checked against the source documents.\" The joint legal conclusions locate the failure in not understanding the inherent risks of AI-based research and drafting tools, in not adequately reviewing citations to the intrinsic record before submission, and in not conducting an inquiry reasonable under the circumstances before certifying the paper to the tribunal.", "reliance_or_harm": "The erroneous citations reached a claim construction chart filed with the district court. Once the parties identified them the respondent reviewed the chart, found further errors, and circulated a corrected version the following day. The order records that the court imposed no sanction and that the respondent represents his client suffered no prejudice and that the underlying suit was later dismissed with prejudice on settlement. The USPTO publicly reprimanded him under 37 C.F.R. §§ 11.101, 11.103, 11.804(c) and 11.804(d), with the order published in the OED FOIA Reading Room and a Notice of Public Reprimand in the Official Gazette.", "public_matter_type": "attorney disciplinary order", "public_matter_name": "In re Brian E. Mitchell, Proceeding No. D2026-16 (USPTO Office of Enrollment and Discipline)", "filing_status": "reprimanded", "jurisdiction": "U.S. Patent and Trademark Office", "public_record_link": "https://foiadocuments.uspto.gov/oed/Mitchell-Order-D2026-16-Redacted.pdf", "confidence_score": "high", "notes_on_resolution": "Final Order approving a proposed settlement agreement, executed by the respondent on 2026-07-20 and by the OED Director on 2026-07-21, approved by the USPTO Director and setting out joint stipulated facts, joint legal conclusions and an agreed-upon sanction. Admitted on the stipulated facts: the order discusses AI in its own voice as the mechanism of the defective citations, so criterion 1 is satisfied without recourse to the undetermined-attribution rule. Admitted notwithstanding the absence of a court sanction, under the outcome-valence rule in INTENT.md: the AI-attributable defect is established even though the district court imposed no penalty. The order names neither AI tool, gives no date for the claim-chart submission, and states no proceeding filing date, so none is recorded here. Source observed 2026-09-09; the retained official PDF and its extracted text are held under data/admission/raw/aiel-2026-069.", "tags": "fabricated-citations; misattributed-quotations; attorney-discipline; uspto; public-reprimand; patent-prosecution; no-court-sanction; legal-services", "source_quality": "primary record", "research_status": "included", "last_checked_date": "2026-09-09", "needs_review": "no", "legal_graph": { "proceedings": [ { "id": "aiel-2026-069-proceeding", "heard_by": [ "u-s-patent-and-trademark-office" ], "determination_ids": [ "aiel-2026-069-determination" ], "procedural_stage": "federal-discipline" } ], "determinations": [ { "id": "aiel-2026-069-determination", "issued_by": [ "u-s-patent-and-trademark-office" ], "issued_date": "2026-07-27", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2026-070", "error_title": "S.D.N.Y. sanctions counsel over ChatGPT-fabricated quotations and imposes a standing GenAI certification duty", "ai_system_name": "ChatGPT (named; admitted)", "deployer": "Brian M. Higbie, Law Offices of Brian M. Higbie, counsel for plaintiff", "domain": "legal services", "error_type": "fabricated quotations and misrepresented holdings", "error_description": "Responding to a sua sponte order to show cause, plaintiff’s counsel acknowledged that he relied on ChatGPT when researching and drafting his opposition papers. The court identified quotations attributed to real Second Circuit decisions that it could not locate in those decisions, together with characterizations the cited cases did not support. Counsel stated that he verified only that each case existed, was in the right jurisdiction and concerned the general subject matter, and did not verify the characterizations before citing them.", "canonical_source_conflicted": "The text and holdings of real Second Circuit decisions, including Back v. Hastings On Hudson Union Free School District, Mangino v. Incorporated Village of Patchogue and Dolan v. Connolly, against quotations and propositions those opinions do not contain", "mitigation_gap": "Counsel had previously encountered a hallucinated citation from the same tool in an unrelated matter and stopped using AI tools, then resumed after an acquaintance in the AI industry told him the fabrication problem had been fixed. His verification step confirmed only that cases existed, never that the quotations were real.", "reliance_or_harm": "The fabricated quotations reached opposition papers on motions to dismiss. The court found a Rule 11 violation and ordered counsel personally to pay $500 to the Clerk of Court, complete at least two New York CLE credits in Ethics and Professionalism on AI, serve the order on his client, and include a certification on any future filing in which he used GenAI describing which portions it assisted, which tools were used and how he confirmed the output. Defendants’ requests for $6,500 and $3,500 in fees were denied because no party had moved under Rule 11, so a sua sponte sanction could only run to the court.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Tiekert v. Village of Mamaroneck, No. 23-CV-8714 (CS) (S.D.N.Y.), ECF No. 100", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Southern District of New York", "filing_date": "2026-09-14", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.nysd.607614/gov.uscourts.nysd.607614.100.0.pdf", "confidence_score": "high", "notes_on_resolution": "Order of Judge Cathy Seibel, 14 September 2026, resolving a sua sponte Rule 11 order to show cause. The forward-looking GenAI certification requirement is the distinguishing feature: it binds counsel on any future filing rather than only in this case, and the corpus holds no comparable standing disclosure remedy. The order places the conduct in the middle of a three-part hallucination taxonomy, fabricated quotations attributed to real decisions rather than wholly nonexistent cases. The denial of opposing counsel’s fee requests turns on the procedural posture, not on the gravity of the conduct.", "tags": "fabricated-quotations; misrepresented-holdings; rule-11; chatgpt; cle-ordered; genai-certification-duty; no-fee-award; legal-services", "source_quality": "primary record", "research_status": "included", "last_checked_date": "2026-09-22", "needs_review": "no", "legal_graph": { "proceedings": [ { "id": "aiel-2026-070-proceeding", "heard_by": [ "u-s-district-court-southern-district-of-new-york" ], "determination_ids": [ "aiel-2026-070-determination" ], "procedural_stage": "trial" } ], "determinations": [ { "id": "aiel-2026-070-determination", "issued_by": [ "u-s-district-court-southern-district-of-new-york" ], "issued_date": "2026-09-14", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2026-071", "error_title": "D.C. Court of Appeals strikes a bank’s brief over nonexistent cases from a generative AI search tool", "ai_system_name": "Google’s generative artificial intelligence search tool (named in the order; no product name given; admitted)", "deployer": "Loishirl W. Hall, then of McCabe, Weisberg & Conway, LLC, counsel for appellee Deutsche Bank National Trust Company", "domain": "legal services", "error_type": "fabricated authority", "error_description": "Ordered to show cause why its brief should not be stricken for citing nonexistent cases, appellee’s counsel responded that she had used Google’s generative artificial intelligence search tool to assist in locating case authority and did not verify the existence or accuracy of the citations before filing. The court found four cited decisions do not exist and declined to reproduce the fabricated citations in its own opinion. The order records that counsel relied on the tool not only to find citations but to draft the brief.", "canonical_source_conflicted": "District of Columbia case law, against four cited decisions that do not exist: Abadie v. District of Columbia, Cason v. National Consumer Co-op Bank, Osborne v. District of Columbia and Woods v. United States", "mitigation_gap": "No verification step stood between a consumer search tool’s generative output and a brief filed on behalf of a major financial institution. Two further attorneys were listed on the brief and the firm responded separately, so the firm’s own review did not catch the fabrications either.", "reliance_or_harm": "The brief was stricken and the matter referred to the District of Columbia Office of Disciplinary Counsel for any investigation it deems appropriate. No monetary sanction was imposed. The court separately referred to its own Rules Committee the question of the scope of its sanctions authority in these circumstances, which the concurring opinion describes as unclear.", "public_matter_type": "appellate order striking a brief with disciplinary referral", "public_matter_name": "Douglas v. Deutsche Bank National Trust Company, No. 24-CV-1099 (D.C. Ct. App.)", "filing_status": "brief stricken; referred for discipline", "jurisdiction": "District of Columbia Court of Appeals", "filing_date": "2026-09-03", "error_date": "2026", "public_record_link": "https://dccourts.gov/sites/default/files/2026-09/Douglas%20v.%20Deutsche%20Bank%20Nat%27l%20Trt%20Co.%2024-CV-1099%20ORDER.pdf", "secondary_source_links": "https://reason.com/volokh/2026/09/04/incredulity-that-competent-law-firm-representing-one-of-the-largest-financial-institutions-in-the-world-filed-brief-with-ai-hallucinations/", "confidence_score": "high", "notes_on_resolution": "Published per curiam order of 3 September 2026 (Easterly and Shanker, Associate Judges; Glickman, Senior Judge, concurring), on responses to a show-cause order. The tool is recorded exactly as the order describes it and is deliberately not resolved to Gemini or AI Overviews, neither of which the order names. This is active use of a consumer search tool for legal research, which distinguishes it from the corpus’s passive-exposure records AIEL-2026-064 and AIEL-GLOB-013, where an AI summary appeared unsolicited in ordinary search results. The concurrence’s discussion of firm-level controls and the Rules Committee referral are both about institutional response rather than this attorney’s conduct.", "tags": "fabricated-citations; consumer-search-ai; active-use; brief-stricken; disciplinary-referral; no-monetary-sanction; appellate; legal-services", "source_quality": "primary record", "research_status": "included", "last_checked_date": "2026-09-22", "needs_review": "no", "legal_graph": { "proceedings": [ { "id": "aiel-2026-071-proceeding", "heard_by": [ "district-of-columbia-court-of-appeals" ], "determination_ids": [ "aiel-2026-071-determination" ], "procedural_stage": "appeal" } ], "determinations": [ { "id": "aiel-2026-071-determination", "issued_by": [ "district-of-columbia-court-of-appeals" ], "issued_date": "2026-09-03", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2026-072", "error_title": "D. Utah penalizes two attorneys over briefing drafted with four generative AI tools", "ai_system_name": "ClearBrief, Claude, ChatGPT and Gemini (named; admitted)", "deployer": "Chad Pehrson (drafting attorney) and Robert S. Clark (signing attorney), counsel for Brigham Young University", "domain": "legal services", "error_type": "fabricated and misattributed authority", "error_description": "Briefing filed for BYU contained errors and citations that appeared to be AI-hallucinated. At a show-cause hearing the drafting attorney stated that he relied on several AI models in drafting the briefing, naming ClearBrief, Claude, ChatGPT and Gemini. Before the show-cause order issued he had filed a notice of errata identifying and correcting twelve errors, and counsel had voluntarily refunded opposing counsel’s fees. The court found a Rule 11(b) violation as to both attorneys on their own admissions.", "canonical_source_conflicted": "The holdings of cited authority, including Oxford Health Plans, which counsel cited for a proposition about evident partiality under 9 U.S.C. section 10(a)(2) that the case does not address", "mitigation_gap": "Four separate AI tools were used in drafting and none of them, nor the signing attorney’s review, caught the fabricated and misattributed authority. The order draws no distinction between ClearBrief, a citation-checking product, and the three general-purpose models, and attributes all four to drafting rather than verification.", "reliance_or_harm": "The court ordered the drafting attorney to complete two CLE courses on the ethical use of AI within six months, and ordered penalties paid to the court of $2,000 from Mr Pehrson and $1,000 from Mr Clark, with proof of payment within thirty days. The refund of opposing counsel’s fees was voluntary and preceded the order, so it is not a fee-shifting sanction.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Beus Gilbert PLLC v. Brigham Young University, No. 2:12-cv-00970-TS (D. Utah), Dkt. 385", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, District of Utah", "filing_date": "2026-09-09", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.utd.86552/gov.uscourts.utd.86552.385.0.pdf", "confidence_score": "high", "notes_on_resolution": "Memorandum decision and order of Judge Ted Stewart, 9 September 2026, following a show-cause order of 30 July 2026 and a hearing on 2 September 2026. The penalties are several and unequal, $2,000 and $1,000 against two named attorneys rather than a single $3,000 obligation, and the CLE requirement binds only the drafting attorney; the signing attorney told the court he had already enrolled in a course himself. This is the corpus’s first record naming four distinct AI tools in one matter, and the only one to date in which a citation-checking product appears alongside general-purpose models without the order distinguishing their roles.", "tags": "fabricated-citations; misattributed-authority; rule-11; multiple-tools; clearbrief; claude; chatgpt; gemini; cle-ordered; legal-services", "source_quality": "primary record", "research_status": "included", "last_checked_date": "2026-09-22", "needs_review": "no", "legal_graph": { "proceedings": [ { "id": "aiel-2026-072-proceeding", "heard_by": [ "u-s-district-court-district-of-utah" ], "determination_ids": [ "aiel-2026-072-determination" ], "procedural_stage": "trial" } ], "determinations": [ { "id": "aiel-2026-072-determination", "issued_by": [ "u-s-district-court-district-of-utah" ], "issued_date": "2026-09-09", "disposition": "confirmed" } ] } }, { "error_id": "AIEL-2026-073", "error_title": "C.D. Illinois fines counsel whose AI-built motion template carried hallucinated citations", "ai_system_name": "Unnamed generative AI (admitted by counsel)", "deployer": "David Baldemar Reyes, counsel for plaintiff Morgan Cole", "domain": "legal services", "error_type": "fabricated authority", "error_description": "Granting default judgment in part on 24 July 2026, the court found the motion contained several hallucinated citations carrying what it called classic indicators of unverified artificial intelligence usage. Responding to the resulting show-cause order, counsel stated that the citations were included because he had prepared a template using AI which inserted them, and accepted full responsibility for the errors.", "canonical_source_conflicted": "Seventh Circuit and district case law on remedies under the Americans with Disabilities Act, against citations the court could not locate", "mitigation_gap": "The defective citations lived in a reusable motion template rather than in a single filing, so the same unverified AI output could recur across matters. Counsel had admitted prior AI hallucinations in another case seven days before filing this motion.", "reliance_or_harm": "The court ordered counsel personally to pay a fine of $1,000 to the Clerk of Court within sixty days, to send a copy of the order to the Illinois Attorney Registration and Disciplinary Commission, and to file a status report on compliance within sixty days. The court described the penalty as deliberately modest because at the time of this filing the only consequence counsel had faced was a stricken motion. No fees or costs were awarded, none having been timely requested.", "public_matter_type": "sanctions order (Rule 11)", "public_matter_name": "Cole v. Hobby Town Unlimited, Inc., No. 4:25-cv-04217-SLD-RLH (C.D. Ill.), ECF No. 33", "filing_status": "sanctioned", "jurisdiction": "U.S. District Court, Central District of Illinois", "filing_date": "2026-09-02", "error_date": "2026", "public_record_link": "https://storage.courtlistener.com/recap/gov.uscourts.ilcd.98203/gov.uscourts.ilcd.98203.33.0.pdf", "confidence_score": "high", "notes_on_resolution": "Order of Chief Judge Sara Darrow, 2 September 2026, the post-judgment sanctions phase following the 24 July 2026 default-judgment order that first identified the citations. The template mechanism is what distinguishes this record: the fabrications were embedded in reusable drafting material rather than produced for one brief. The order cross-references two other matters involving the same attorney, Hippe v. Retention Brands, No. 2:25-cv-00596 (E.D. Wis.) and Pittman v. Hazel & Olive, No. 1:25-cv-07791 (N.D. Ill.), neither of which is in this corpus. The docket’s CourtListener title reflects the stale original caption naming a terminated co-defendant.", "tags": "fabricated-citations; reusable-template; rule-11; attribution-admitted; ardc-referral; repeat-conduct; legal-services", "source_quality": "primary record", "research_status": "included", "last_checked_date": "2026-09-22", "needs_review": "no", "legal_graph": { "proceedings": [ { "id": "aiel-2026-073-proceeding", "heard_by": [ "u-s-district-court-central-district-of-illinois" ], "determination_ids": [ "aiel-2026-073-determination" ], "procedural_stage": "trial" } ], "determinations": [ { "id": "aiel-2026-073-determination", "issued_by": [ "u-s-district-court-central-district-of-illinois" ], "issued_date": "2026-09-02", "disposition": "confirmed" } ] } } ] }, "review": { "label": "Review queue", "description": "Borderline or not-yet-verified candidates requiring primary-record capture or scope decisions.", "records": [ { "candidate_id": "AIEL-CAND-001", "candidate_title": "RealPage revenue-management algorithmic rent-setting", "candidate_matter": "United States et al. v. RealPage, Inc.", "reason_for_review": "Strong public matter, but antitrust price coordination may be adjacent to the error-event scope rather than a discrete AI error; proposed consent decree status also needs refresh.", "next_verification_step": "Fetch complaint and proposed consent decree; decide whether row should be tagged as algorithmic pricing harm or excluded as non-error antitrust conduct.", "best_available_sources": "https://justice.gov/archives/opa/pr/justice-department-sues-realpage-algorithmic-pricing-scheme-harms-millions-american-renters; https://justice.gov/opa/pr/justice-department-requires-realpage-end-sharing-competitively-sensitive-information-and", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-002", "candidate_title": "UnitedHealthcare nH Predict post-acute-care denials", "candidate_matter": "Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc. et al.", "reason_for_review": "Specific AI system alleged, but the record is an active class action and each denial may be a separate event; primary complaint/order should be archived before inclusion.", "next_verification_step": "Retrieve complaint, dismissal order, and scheduling order from court or Georgetown tracker; decide whether to create one representative row or event-level rows.", "best_available_sources": "https://litigationtracker.law.georgetown.edu/litigation/estate-of-gene-b-lokken-the-et-al-v-unitedhealth-group-inc-et-al/; https://cbsnews.com/news/health-insurance-humana-united-health-ai-algorithm/", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-003", "candidate_title": "Cigna PXDX bulk automated claim denials", "candidate_matter": "Kisting-Leung et al. v. Cigna Corp. et al.", "reason_for_review": "Strong candidate for health-insurance automation, but primary order/complaint should be obtained and the error may need splitting by claim-denial category.", "next_verification_step": "Fetch complaint and March 2025 order; verify whether California statutory physician-review theory has a final or surviving claim status.", "best_available_sources": "https://courthousenews.com/judge-advances-class-claims-over-cigna-use-of-automated-algorithm-to-deny-benefits/; https://topclassactions.com/lawsuit-settlements/insurance/cigna-class-action-claims-software-automatically-denies-large-batches-of-claims-without-review/", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-004", "candidate_title": "Oready AI-hallucinated GAO bid protests", "candidate_matter": "Oready, LLC, B-423524.2; B-423649; B-423650; B-423670; B-423670.2", "reason_for_review": "Primary GAO decisions are verified. B-423524.2 found repeated nonexistent or materially false citations and said the errors bore the hallmarks of unverified large-language-model or other AI use. The later consolidated decision dismissed four protests as an abuse of GAO's bid-protest process after the conduct continued. Attribution remains inferential rather than an admission or unhedged finding that AI was used, so promotion depends on the corpus's gate-1 policy.", "next_verification_step": "Make the steward call on whether GAO's express 'hallmarks' finding satisfies gate 1. If it does, draft the matter from the two verified GAO decisions and preserve the distinction between the initial warning, the second warning, and the later abuse-of-process dismissal.", "best_available_sources": "https://gao.gov/products/b-423524.2; https://gao.gov/assets/890/881816.pdf", "last_checked_date": "2026-07-25" }, { "candidate_id": "AIEL-CAND-005", "candidate_title": "Whiting appellate fake citations and sanctions", "candidate_matter": "Whiting v. City of Athens, Tennessee", "reason_for_review": "Large sanctions event, but AI use appears suspected rather than admitted or expressly found; primary opinion may be behind Westlaw.", "next_verification_step": "Retrieve official Sixth Circuit opinion if published; confirm whether the court made an AI finding or only analogized to hallucinations.", "best_available_sources": "https://sixthcircuitappellateblog.com/recent-cases/sixth-circuit-sanctions-attorneys-for-fake-citations-what-does-this-mean-for-use-of-ai/; https://jdsupra.com/legalnews/6th-circuit-court-of-appeals-issues-six-5453130/", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-006", "candidate_title": "Couvrette winery inheritance fake citations sanctions", "candidate_matter": "Couvrette v. Wisnovsky", "reason_for_review": "Potentially important client-supplied AI citation case, but primary orders are Westlaw/PACER and the AI-origin finding needs exact language.", "next_verification_step": "Retrieve December 2025 and April 2026 D. Or. orders; verify whether client AI generation is finding, inference, or allegation.", "best_available_sources": "https://nwsidebar.wsba.org/2026/03/02/parade-of-horribles-federal-court-in-oregon-surveys-sanctions-for-ai-fake-citations/; https://abajournal.com/news/article/oregon-federal-judge-hands-down-110000-penalty-for-ai-errors", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-007", "candidate_title": "Michael Oliver Detroit FRT false match", "candidate_matter": "Oliver v. City of Detroit / related Detroit FRT litigation", "reason_for_review": "Fact pattern appears strong, but the current source set is secondary-heavy and the settlement linkage to Williams reforms should be verified from court records.", "next_verification_step": "Locate complaint, settlement, or docket materials; verify vendor, matter name, and procedural outcome.", "best_available_sources": "https://aclumich.org/cases/facial-recognition/; https://aiaaic.org/aiaaic-repository/ai-algorithmic-and-automation-incidents/michael-oliver-facial-recognition-wrongful-arrest", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-008", "candidate_title": "Alonzo Sawyer MTA Police FRT wrongful arrest", "candidate_matter": "State criminal case / no known civil suit", "reason_for_review": "Publicly documented wrongful arrest, but no civil complaint or administrative matter was identified; public legal record needs confirmation.", "next_verification_step": "Obtain criminal docket or dismissal record; decide whether dismissed criminal prosecution is enough to include without a civil matter.", "best_available_sources": "https://wired.com/story/face-recognition-software-led-to-his-arrest-it-was-dead-wrong/; https://aclu.org/news/privacy-technology/more-than-a-dozen-wrongful-arrests-due-to-police-reliance-on-facial-recognition-technology", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-009", "candidate_title": "Jason Killinger Reno casino FRT arrest", "candidate_matter": "Killinger v. City of Reno / Officer Jager", "reason_for_review": "Interesting private-casino FRT fact pattern, but primary complaint and current status should be obtained; date/status references may have changed.", "next_verification_step": "Fetch federal complaint and amended complaint; verify Peppermill settlement and city/RPD status.", "best_available_sources": "https://biometricupdate.com/202604/lawsuit-alleges-systemic-misuse-of-facial-recognition-by-reno-police; https://incidentdatabase.ai/entities/jason-killinger/", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-010", "candidate_title": "Kimberlee Williams Crimedex third-party FRT wrongful arrest", "candidate_matter": "ACLU administrative complaints to Maryland police departments", "reason_for_review": "Compelling factual record, but no lawsuit and civil limitations reportedly expired; decide whether administrative complaints qualify as public matter.", "next_verification_step": "Fetch ACLU complaint letters; verify warrant/docket records and whether administrative complaint category should be included.", "best_available_sources": "https://aclu.org/cases/kimberlee-williams-wrongful-arrest; https://aclu.org/press-releases/woman-wrongly-jailed-for-months-based-on-faulty-facial-recognition-technology-demands-apology-from-maryland-police-departments", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-011", "candidate_title": "TennCare Connect / Deloitte eligibility system denials", "candidate_matter": "Tennessee Medicaid class action", "reason_for_review": "Strong government-benefits candidate, but not yet researched to primary-source standard for this seed release.", "next_verification_step": "Fetch federal court order, complaint, and state system documents; compare to Arkansas RUGs and MiDAS to avoid duplicate-style overrepresentation.", "best_available_sources": "https://gizmodo.com/judge-rules-400-million-algorithmic-system-illegally-denied-thousands-of-peoples-medicaid-benefits-2000492529", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-012", "candidate_title": "Apple Card / Goldman Sachs credit-limit gender-bias investigation", "candidate_matter": "NYDFS Apple Card investigation", "reason_for_review": "Reviewed as a negative or excluded example because the regulator concluded no fair-lending violation; may be useful for exclusions appendix, not main log.", "next_verification_step": "Archive NYDFS report as excluded-case precedent and note why investigation closure without finding is not enough.", "best_available_sources": "https://dfs.ny.gov/reports_and_publications/press_releases/pr202103231", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-CAND-013", "candidate_title": "Pro se litigant sanctioned $1,000 for fabricated case citations", "candidate_matter": "William C. Maxwell v. Dakota Michael, Officer, et al., No. 1:24-cv-01365-JRS-TAB (S.D. Ind.)", "reason_for_review": "Terminal disposition, decided 2026-08-19: **not admissible**, and not to be re-triaged. Under the undetermined-attribution rule in INTENT.md, AI must remain a live and undisplaced explanation, and here the primary source establishes a competing non-AI mechanism. The court did agree, in its own voice, that the \"fabricated authority contains the hallmarks of generative AI hallucinations,\" which would ordinarily satisfy the refined criterion 1. But it went on to hold that \"[w]hether or not Plaintiff used AI, his conduct remains sanctionable, as it represents a deliberate intent to fabricate citations and quotations,\" and recorded the litigant's own account that where he could not find a case with his desired holding he composed fake citations himself as \"placeholders,\" fabricating name, case number, citation, pin cite and court across at least 37 instances. Deliberate human fabrication is the operative finding; the AI hallmarks are displaced by it. Fabrication that merely resembles AI output is ordinary litigation misconduct. Verified against the primary order: Doc 206, filed 2026-05-26 (Mag. J. Baker), $1,000 Rule 11 sanction plus $886.80 deposition costs, resolved through RECAP.", "next_verification_step": "None. This candidate is closed on the merits, not on sourcing. Retained as the corpus's worked example of the competing-mechanism bound, so that a future sweep does not re-admit it on the strength of the court's hallmarks language alone. Do not re-triage.", "best_available_sources": "https://storage.courtlistener.com/recap/gov.uscourts.insd.216295/gov.uscourts.insd.216295.206.0.pdf", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-015", "candidate_title": "Appellate reversal: trial court relied on fictitious case from counsel brief", "candidate_matter": "H. C., a Minor v. Rudy Contreras, No. F089316 (Cal. Ct. App., 5th Dist.) (Super. Ct. No. 23CEFL03867)", "reason_for_review": "Terminal disposition, decided 2026-08-19: **not admissible**, and not to be re-triaged. Under the undetermined-attribution rule in INTENT.md the AI discussion must appear in the primary source; a third party's inference is never a substitute. This opinion never mentions artificial intelligence and attributes the fabricated citation to no tool. The tracker's \"Implied\" tag is the only AI link and it carries no evidentiary weight. The underlying facts remain notable and are recorded here rather than dropped: the unpublished opinion (filed 2026-05-28; Snauffer, J.) reverses and remands a DVPA restraining-order denial because the trial court copied a fictitious citation, Enrique M. v. Angelina V. (2005) 15 Cal.App.5th 788, from respondent counsel's closing brief into its own ruling, and reassigns the matter to a different judge. That is proposed-order contamination reaching a judicial document, which is a pattern the corpus tracks — but without any AI reference in the source it is a fabricated-citation matter, not an AI incident.", "next_verification_step": "None. Closed on the merits. Re-open only if a later published decision or disciplinary proceeding in the same matter introduces an AI attribution into the record. Do not re-triage.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-016", "candidate_title": "Pro se appellants’ briefs struck over hallucinated citations (Centient AI referenced)", "candidate_matter": "Strong v. Zubha Pop Foods LLC d/b/a Popeyes, 2026 IL App (1st) 242451-U", "reason_for_review": "Gate 1 FAIL (2026-06-02 re-verification). Court language is hedged throughout: 'note the possibility', 'apparent unchecked use', 'apparent reliance on AI'. No actor admission (pro se, no disclosure). Court expressly declined to sanction. Briefs struck solely for Rule 341 noncompliance. Centient AI appears in plaintiffs' own briefs as a hyperlink reference — circumstantially strong but not an admission and not adopted as a court finding. Cannot promote to included under strict gate-1 standard. Retain for future promotion if gate-1 standard relaxes to include strong-inference court language, or if plaintiffs' briefs are obtained as a corroborating source.", "next_verification_step": "Pull plaintiffs' original briefs to confirm Centient AI product reference is unambiguous; then decide whether to relax gate-1 to cover strong-inference AI findings for pro se litigants.", "best_available_sources": "", "last_checked_date": "2026-06-02" }, { "candidate_id": "AIEL-CAND-017", "candidate_title": "W.D. Pa. declines to sanction pro se litigant’s disclosed use of Claude for drafting; no fabricated citations found", "candidate_matter": "Mullins v. Duquesne University of the Holy Spirit, Civil No. 25-1366 (W.D. Pa.), 2026 WL 2042802", "reason_for_review": "Terminal disposition, decided 2026-08-19: this matter is **not admissible** and is not expected to be promoted. Gate 1 passes (the order names Anthropic's Claude, disclosed by the pro se plaintiff under the court's standing AI-use order and credited as fact by the court), but under the outcome-valence rule added to INTENT.md the corpus admits a matter on an established AI-attributable defect, and no defect was established here. The court found the brief presented \"neither inaccurate, false, or non-existent case citations\"; the alleged defects were real cases cited for propositions they do not support, that is mis-citation rather than fabrication. It denied the motion to strike and held the conduct did not amount to sanctionable AI use, noting it found no cases imposing sanctions for using AI only to assist with drafting and editing. Disclosed AI use plus an allegation that did not survive scrutiny is not an incident. The matter retains real value as a limiting precedent on what AI use is *not* sanctionable, which is why it is retained rather than dropped, but that value belongs in the methodology narrative and not in the incident dataset.", "next_verification_step": "None for admission purposes. Record the matter in `docs/methodology.html` as the worked example of the rule's exclusionary half, alongside In re Bard as the worked example of its inclusionary half. Re-open only if a later order in the same proceeding establishes an actual AI-attributable defect. Do not re-triage this candidate on future sweeps.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-018", "candidate_title": "South African Labour Court stays disciplinary hearing over hearing chairperson’s non-existent citations; AI source left undetermined", "candidate_matter": "M J Molawa and Others v Matjhabeng Local Municipality and Another, Case No 2026-157576 (Labour Court of South Africa, Durban)", "reason_for_review": "Gate 1 FAILS on attribution, not on fabrication. The court confirmed that authorities cited in the chairperson’s 3 June ruling do not exist, but the applicants only pleaded AI as the source and the court expressly declined to decide it, framing the point conditionally and holding the outcome identical either way: whether the citations were AI-generated or self-generated, there is prima facie evidence the chairperson failed to apply his mind to the law. No AI product is named, no admission, no referral for discipline, no costs. Two further cautions: the tracker tags this as judge-side AI use, but the alleged user is the private chairperson of an internal municipal disciplinary hearing under a SALGBC collective agreement, not a judicial officer or a bargaining-council arbitrator; and the linked PDF is a third-party copy of the court’s own Word-to-PDF hand-down, not a SAFLII version. Non-US, so this would be a global candidate if the AI finding ever materialises.", "next_verification_step": "Watch the separate pending review application, Case No 2026-152359, which seeks to set aside the chairperson’s rulings and may determine the source of the citations; obtain an official SAFLII copy of the present judgment in the meantime.", "best_available_sources": "", "last_checked_date": "2026-07-24" }, { "candidate_id": "AIEL-CAND-019", "candidate_title": "Delhi High Court stays a commercial-court decree over a judge’s own suspected AI-drafted reasoning, without deciding whether AI was used", "candidate_matter": "SNV Aviation Pvt Ltd v. ABS Tour and Travels, RFA(COMM) 284/2026 (High Court of Delhi), on appeal from CS(Comm) 322/2024 (District Judge (Commercial)-02, South-East, Saket Courts)", "reason_for_review": "Gate 1 fails on the court’s own express reservation. The High Court recorded that the manner of drafting in the impugned judgment “prima facie, gives an impression that there is some AI softwares which may have been used”, but immediately held that “it cannot be conclusively said until further examination as to whether any AI software has been used by the ld. District Judge”. That is a prima facie impression coupled with an explicit refusal to find, not an establishment of AI use, and no product is named. Retained because it is the highest-value class of lead the corpus wants — adjudicator-side, where the defect sits in the judge’s own judgment rather than in counsel’s filings — and because it is live: the trial court record has been requisitioned and the matter listed for hearing, so the impression may yet become a finding. Note the tracker’s forum is wrong: the operative order is the High Court of Delhi, not the Saket commercial court whose judgment is impugned, and the alleged defect is non-existent propositions attributed to genuine Supreme Court authorities rather than invented case names. The Rs 20 lakh deposit is a condition of the stay, not a penalty. Any future record must describe this as an interim order in a pending appeal, not a resolved matter.", "next_verification_step": "Re-check after the 2026-08-20 hearing to see whether the High Court converts its prima facie impression into a finding of AI use; separately obtain the appended Saket commercial court judgment of 2026-02-24 from the eCourts record for CS(Comm) 322/2024, since the official High Court URL serves only the three-page appellate order.", "best_available_sources": "https://delhihighcourt.nic.in/app/showFile/1777714409_42b8687818fac18a_pms_2842026.pdf/2026", "last_checked_date": "2026-07-24" }, { "candidate_id": "AIEL-CAND-020", "candidate_title": "Georgia Court of Appeals vacates an order containing a hallucinated case that counsel drafted and the trial court signed", "candidate_matter": "Henry County Schools et al. v. Grant / Douglas / Forbes, Nos. A26A0698, A26A0699, A26A0700 (Ga. Ct. App., 1st Div.)", "reason_for_review": "Gate 1 is arguable rather than clear, and gate 2 currently has no citable link. On the merits the Court of Appeals found the trial court’s order “relies on at least one hallucinated case” and, quoting Slay v. Ross that hallucinated cases are typically a hallmark of irresponsible use of generative AI, went on to “remind counsel, and lower courts, of their professional responsibilities and obligations in the use of artificial intelligence”. That is court-side AI attribution addressed to these parties, but the court never states that counsel used AI and hedges with “typically”, so the finding is by implication rather than fact-finding. The fabricated authority sat in the trial court’s signed order, but the opinion says that order was drafted by the plaintiffs’ attorney — so it is counsel-side conduct carried into a judicial document, not a judge independently using AI. No sanction, referral, or costs issued. Vacatur rested partly on unrelated grounds, so the hallucinated case is a contributing rather than sole basis. Sourcing problem: the entire gaappeals.gov publication surface sits behind a CAPTCHA gate that returns HTTP 405 to every programmatic request, so no court-published URL could be verified, and the corpus no longer cites third-party rehosts. The matter is fully identified by its official caption and docket numbers without a URL.", "next_verification_step": "Retrieve the court’s own opinion PDF for A26A0698 from the gaappeals.gov opinion search in a browser and record that URL; capture Slay v. Ross, 927 SE2d 569 (Ga. App. 2026) at the same time, since it supplies the AI language and is a separate candidate. Then make a scope call on whether attribution by implication satisfies gate 1.", "best_available_sources": "", "last_checked_date": "2026-07-24" }, { "candidate_id": "AIEL-CAND-021", "candidate_title": "Quebec Superior Court annuls an arbitral award because the arbitrator’s own reasoning rested entirely on hallucinated authorities", "candidate_matter": "ARIHQ et Centre de Santé Osman c. Santé Québec (CCSMTL) et Me Michel A. Jeanniot, 2026 QCCS 1360, Nos 500-11-066427-259 and 500-11-066498-250 (Cour supérieure du Québec, Chambre commerciale)", "reason_for_review": "Gate 1 passes on the merits; gate 2 has no reachable original source, which is the only reason this is not in the global bucket. On the merits this is the strongest adjudicator-side matter found in the 2026 sweep and the corpus’s first private-arbitration annulment: the court found that all of the doctrinal and case authorities the arbitrator relied on were non-existent and “hallucinées”, that the arbitrator’s authority had been delegated and he had abdicated his role of reviewing the result, and annulled the award under arts. 646 and 648 C.p.c. for failure to respect the arbitral procedure. AI delegation was the sole successful ground: the public-order ground was expressly rejected and a natural-justice ground was withdrawn. The court also rejected the argument that it should refrain from finding AI use absent expert evidence, holding the civil balance-of-probabilities standard sufficient. Five defective authorities were identified, including one SOQUIJ confirmed by letter does not exist. No product is named, so the establishment basis is a judicial finding rather than naming. Sourcing problem: CanLII returns HTTP 403 behind a CAPTCHA wall, SOQUIJ refuses connections outright, and the Quebec court portals sit behind a JavaScript challenge, so no original-source URL has been verified and the corpus no longer cites the tracker or its storage bucket.", "next_verification_step": "Confirm the CanLII address for 2026 QCCS 1360 in a browser, or obtain a CanLII API key, then promote to the global bucket with the verified link. Do not let law-firm commentary set the characterisation of the holding — at least one firm hedges the AI finding as merely probable, which the judgment itself does not.", "best_available_sources": "", "last_checked_date": "2026-07-24" }, { "candidate_id": "AIEL-CAND-024", "candidate_title": "GAO warns protester after admitted reliance on AI-assisted tools produced citation irregularities", "candidate_matter": "Raven Investigations & Security Consulting, LLC, B-423447 (U.S. Government Accountability Office)", "reason_for_review": "The published GAO decision confirms that the protester's non-attorney representative relied on AI-assisted tools, public databases, online repositories, and secondary legal-research summaries, and that the identified citation irregularities arose from those automated or secondary tools. GAO found nonexistent, mismatched, and unsupported authorities and quotations. It dismissed the protest as academic for an independent reason, declined to impose sanctions in this matter, but expressly warned that future filings with nonexistent authority may result in sanctions. The remaining policy seam is narrower than the 'hallmarks' candidates: AI use is admitted, but the representative did not isolate AI as the exclusive cause among the several tools used.", "next_verification_step": "Steward call on whether admitted AI-assisted work within a mixed-tool causal account satisfies gate 1. If yes, promote as a formal administrative-adjudication warning, not as a sanction.", "best_available_sources": "https://gao.gov/products/b-423447", "last_checked_date": "2026-07-25" }, { "candidate_id": "AIEL-CAND-025", "candidate_title": "New York court sanctions counsel for deceptive litigation and orders disclosure about possible AI use", "candidate_matter": "Facey v. Fisher, 2025 NY Slip Op 33456(U), Index No. 152088/2025, NYSCEF Doc. No. 50 (N.Y. Sup. Ct., New York County)", "reason_for_review": "The official opinion verifies multiple nonexistent or materially miscited authorities and awards defendants their legal fees against plaintiff's counsel, but the sanctions also rest on separate deceptive litigation conduct. Gate 1 is not yet established: the court ordered counsel to file an affirmation within five days stating whether artificial-intelligence applications or chatbots were used and explaining the citation failures. The opinion itself asks the question and does not answer it.", "next_verification_step": "Retrieve counsel's post-order NYSCEF affirmation and any later fee or sanctions order. NYSCEF guest case search for Index No. 152088/2025 was rechecked on 2026-07-25 but placed the document list behind hCaptcha; ordinary web indexing exposed only NYSCEF Doc. No. 50. A human browser pass through the NYSCEF guest-search challenge is therefore the next source step. Promote only if the follow-up record establishes AI use and permits the AI-attributable conduct and outcome to be separated from the independent misconduct.", "best_available_sources": "https://nycourts.gov/reporter/pdfs/2025/2025_33456.pdf", "last_checked_date": "2026-07-25" }, { "candidate_id": "AIEL-CAND-026", "candidate_title": "New York court imposes maximum statutory sanction plus $36,511 fees on attorney who used Claude for nineteen filings", "candidate_matter": "Kleyman Law Group, P.C. v. Kaloidis, No. 502644/2025, 2026 WL 2324499 (N.Y. Sup. Ct., Kings County, Aug. 4, 2026) (Capell, J.), Decision and Order after Hearing", "reason_for_review": "Gate 1 passes in its strongest form: at the sanctions hearing Alex I. Kleyman testified that from April to December 2025 he used \"Claude\" and other generative-AI research platforms to prepare the submissions, that he could not afford Lexis or Westlaw, and that he later used the same platforms to verify citations after opposing counsel questioned them. The court found nineteen separate submissions collectively citing twenty-three fictitious cases on forty-seven occasions and misstating genuine authority at least eighty-three times, conduct it said \"seem[s] to exceed most, if not all, published cases on this topic to date.\" Gate 2 fails for sourcing only: the order is available to this project solely as a Westlaw slip copy. NYSCEF returns HTTP 403 to scripted requests, and the matter carries no New York Official Reports slip-opinion citation, so no court-hosted original was reachable and no link is recorded.", "next_verification_step": "Retrieve the order from NYSCEF for Kings County Index No. 502644/2025 through a human browser pass (the order directs proof of payment to be filed to NYSCEF, so the docket is public), or wait for a New York Official Reports posting. Note for drafting: the tracker records this as a single \"46511 USD\" monetary sanction, which is wrong in both character and obligor count. The order imposes two distinct payments, each joint and several against Kleyman Law Group, P.C. and Alex I. Kleyman personally: a $10,000 sanction to the Lawyer's Fund for Client Protection, being the maximum permitted for a single occurrence under 22 N.Y.C.R.R. 130-1.2, and $36,511.35 in reasonable attorneys' fees and costs to defence counsel Wenig Saltiel LLP. The court separately referred Kleyman to the Grievance Committee for the Second, Eleventh and Thirteenth Judicial Districts.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-027", "candidate_title": "New York Family Court sanctions counsel $1,000 for eleven fabricated citations but declines grievance referral", "candidate_matter": "Mendel S. v. Faigy S., File No. 186725, 2026 NY Slip Op 26118, 2026 WL 2264143 (N.Y. Fam. Ct., Kings County, July 31, 2026) (Paley, J.)", "reason_for_review": "Gate 1 passes on a court finding without a named product: the court found that counsel repeatedly cited hallucinated cases generated by generative artificial intelligence which he did not verify, and that eleven of sixteen citations were fabricated. No AI product is identified anywhere in the operative findings. The order does contain the word \"Copilot\", but only inside a parenthetical quotation from Idehen v. Stoute-Phillip, 86 Misc 3d 1244(A); it is not a finding about this attorney and must not be recorded as a named tool. Gate 2 fails for sourcing only: the New York Official Reports Miscellaneous 3d slip-opinion index at nycourts.gov currently reaches only 2026_26042, so slip opinion 26118 is not yet posted, and both the .htm and .pdf reporter paths return the site's soft-404 page under HTTP 200.", "next_verification_step": "Re-check https://nycourts.gov/reporter/3dseries/2026/2026_26118.htm in a few weeks; the Official Reports posting lags Westlaw and this citation should appear. Confirm on retrieval that the sanction is $1,000 payable to the Lawyers' Fund for Client Protection under 22 NYCRR 130-1.3 with proof of payment filed within ten days, and that the court expressly declined to refer counsel to the Attorney Grievance Committee. Counsel is pseudonymised in the published decision as \"Connor C., Esq.\"; retain the pseudonym in any record.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-028", "candidate_title": "Washington appellate court surmises AI use behind a fabricated citation without making a finding", "candidate_matter": "In re Marriage of Muriel and Farris, No. 60903-7-II (Wash. Ct. App. Div. II, Aug. 18, 2026) (Lee, J.), unpublished opinion", "reason_for_review": "Admissible in principle as of 2026-08-19; **blocked on gate 2 sourcing only**. Under the undetermined-attribution rule in INTENT.md, criterion 1 is satisfied because the court discusses AI in its own voice in connection with the defect: \"We surmise that Jonathan used artificial intelligence (AI) to assist in writing his brief, which resulted in an 'AI hallucination.'\" A surmise is not a finding, but the rule no longer requires one, and no competing non-AI mechanism is established in the opinion. The court confirmed the cited case does not exist and considered sanctions for a frivolous appeal under RAP 18.9(a). Gate 2 fails: courts.wa.gov did not resolve from this network and a CourtListener opinion search for the parties in washctapp after 2026-08-01 returned zero results, so no court-hosted original was reachable.", "next_verification_step": "Retrieve the opinion from courts.wa.gov (Division II, No. 60903-7-II, filed 18 August 2026) and promote to `included` with the qualifier `(raised in the record; attribution not determined)`, the tag `attribution-not-determined`, `confidence_score` no higher than `medium`, and the court's surmise quoted in `notes_on_resolution`.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-029", "candidate_title": "Arizona MDL court declines to exclude expert report despite hallucinated citations and conceded ChatGPT use", "candidate_matter": "In re Bard Implanted Port Catheter Products Liability Litigation, MDL No. 3081 (D. Ariz., Aug. 12, 2026), Order re Ahmed El-Ghannam, Ph.D., 2026 WL 2333723", "reason_for_review": "Gate 1 passes and the outcome-valence question is now settled in this candidate's favour. Under the rule added to INTENT.md on 2026-08-19, a matter is admitted on the established AI-attributable defect rather than on the severity of the consequence, and this record clears that bar: hallucinated citations were present in Dr. Ahmed El-Ghannam's general causation report, in portions later withdrawn, and he was questioned at deposition about whether he used an artificial-intelligence platform including ChatGPT. The court found his explanation adequate and denied the motion to exclude, noting that other courts have admitted reports in comparable circumstances. The non-exclusion is therefore this record's outcome, not a reason to withhold it. **Only gate 2 now blocks promotion.** The order is available to this project solely as a Westlaw slip copy; a CourtListener RECAP document sweep of docket 2:23-md-03081 in azd returned nineteen deduplicated entries, none of them the 12 August 2026 order and none marked available.", "next_verification_step": "Resolve the order through PACER for D. Ariz. MDL No. 3081 by its 12 August 2026 filing date, or watch for a RECAP upload. On retrieval, promote to `included` with `filing_status` recording that exclusion was denied and `reliance_or_harm` recording that the report was admitted notwithstanding the hallucinated citations. This will be the corpus's first record of a tribunal declining to attach a consequence to an established AI defect, and it pairs with AIEL-2026-062, where hallucinated citations in an expert report produced exclusion and dismissal with prejudice.", "best_available_sources": "", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-CAND-030", "candidate_title": "New Mexico Supreme Court holds attorney in contempt over ChatGPT-fabricated appellate summary", "candidate_matter": "State of New Mexico v. Oscar Renee Sandoval, No. S-1-SC-40845 (N.M. Sup. Ct.) — show-cause proceeding against attorney Stephen Aarons", "reason_for_review": "Facts clear gate 1 decisively (ChatGPT named and admitted on the record; court finding of direct contempt; $5,000 fine; removal from the case; briefs stricken; Disciplinary Board referral) and would otherwise be a strong `included` candidate, but the only public record located is the court's own audio/video recording of the August 21, 2026 show-cause hearing and its oral ruling — no written order has been posted publicly anywhere, including nmcourts.gov, nmonesource.com, or CourtListener (which carries no New Mexico state coverage at all). The corpus's existing `included` records all cite a text document; this one would be the first cited to an audio/video recording, which is a format decision for the steward rather than a sourcing gap.", "next_verification_step": "Check nmcourts.gov and the New Mexico Supreme Court's own case-information system periodically for the written contempt order once it is docketed and posted; if a text order becomes available, promote to `included` using it as public_record_link. Venue detail (\"Las Cruces area\"/Chaparral/Doña Ana County) is press-sourced only (Santa Fe New Mexican, ABA Journal) and should be confirmed against the eventual written order rather than carried forward from news coverage.", "best_available_sources": "https://supremecourt.nmcourts.gov/about-this-court/recordings-of-oral-arguments/; https://supremecourt.nmcourts.gov/wp-content/uploads/sites/2/2026/08/08212026.mp3; https://santafenewmexican.com/news/local_news/new-mexico-high-court-blasts-lawyer-takes-him-off-case-for-ai-errors-in-brief/article_a162daf7-95a4-4010-a017-dedcbdb68b0b.html; https://abajournal.com/news/article/criminal-defense-attorney-admits-stupidity-over-ai-errors-but-still-receives-sharp-rebuke-from-his-states-high-court", "last_checked_date": "2026-09-02" }, { "candidate_id": "AIEL-CAND-032", "candidate_title": "USPTO publicly reprimands trademark attorney for nonexistent and misquoted TTAB citations; order records denial of AI drafting and requires generative-AI CLE", "candidate_matter": "In re Anthony Matos, Proceeding No. D2025-13 (USPTO OED final order, March 6, 2025)", "reason_for_review": "Criteria 2 through 4 are satisfied on the official USPTO final order: the stipulated facts establish that the respondent's September 17, 2023 TTAB trial brief in Cancellation No. 92077944 cited nonexistent decisions (E.J. Brach Corp. v. Gilbert/Robinson; CEP Holdings v. Bieber; U.S. Olympic Committee v. S. Industries) and attributed quotations to real decisions that contain no such language; the USPTO publicly reprimanded him under 37 C.F.R. sections 11.101, 11.103, 11.301, 11.804(c), and 11.804(d) and ordered two hours of CLE on the use of generative AI in legal practice. Criterion 1 is not satisfied on the current reading. The order mentions AI only twice: the respondent asserts he used a generative AI tool to learn TTAB mechanics but did not use any such tool in drafting the brief, and the sanction requires generative-AI CLE. The order records, and the Notice of Public Reprimand adopts in OED's own voice, a non-AI mechanism: the respondent copied citations from unread blog posts and articles found by internet search. Under the 2026-08-19 competing-mechanism bound, AI is displaced as an explanation unless the steward treats the unexplained CLE topic as the tribunal discussing AI in connection with the defect. The order does not name any AI tool and makes no finding that AI produced the citations.", "next_verification_step": "Steward policy decision: decline, or retain in review as a limiting precedent alongside Mullins v. Duquesne and Maxwell v. Michael. Do not promote on this record without a documented reason why an unexplained generative-AI CLE sanction counts as the source discussing AI in connection with the defect despite the recorded denial and the adopted internet-search mechanism. If pursuing admission, first retrieve the TTAB opinion of February 7, 2024 in Cancellation No. 92077944 (TTABVUE) and check whether the Board itself discussed AI; the OED order's quotations from it do not. Any admitted record must carry attribution-not-determined, confidence_score no higher than low, and must not assert that AI caused the defect.", "best_available_sources": "https://foiadocuments.uspto.gov/oed/Matos-Final-Order-(D2025-13)-Redacted.pdf", "last_checked_date": "2026-09-10" }, { "candidate_id": "AIEL-CAND-033", "candidate_title": "Minnesota Court of Appeals affirms sanctions against a self-represented appellant who admitted using AI to draft filings", "candidate_matter": "Asiedu v. Eastlund, Nos. A25-2106 and A26-0xxx (Minn. Ct. App., nonprecedential opinion filed 8 September 2026)", "reason_for_review": "Gate 1 is satisfied on the opinion’s own text: at a hearing the appellant \"admitted that he used artificial intelligence as a research tool to draft his\" filings and did not ensure the accuracy of the citations, and the court affirmed dismissal together with inherent-authority sanctions for citing misleading, misrepresented and non-existent caselaw. The tool is unnamed. The award of $2,341.25 is respondents’ reasonable costs and attorney fees, a fee-shift computed from actual costs rather than a flat fine; the tracker’s \"2341\" drops the cents and implies a penalty. Held on gate 2 alone: no citable original was reachable. The Minnesota opinion archive at mn.gov/law-library-stat sits behind a Radware Bot Manager CAPTCHA that answers HTTP 200 with a challenge page, mncourts.gov returns HTTP 403 to scripted requests, and CourtListener carries no opinion for this matter.", "next_verification_step": "Retrieve the nonprecedential opinion from the Minnesota Court of Appeals through a browser pass at mn.gov/law-library-stat/archive/ctapun/2026/, or via a CourtListener opinion result once one appears, and confirm the second appeal number, the panel and the exact $2,341.25 award before promotion.", "best_available_sources": "", "last_checked_date": "2026-09-22" }, { "candidate_id": "AIEL-CAND-034", "candidate_title": "Missouri Court of Appeals sanctions counsel over two non-existent cases while disclaiming any sanction for using AI as such", "candidate_matter": "State of Missouri ex rel. Steven Schnell v. Honorable Rebecca Richardson, No. ED114731 (Mo. Ct. App. E.D., Writ Division Four, 8 September 2026)", "reason_for_review": "Gate 1 is satisfied on the order’s own text: counsel’s \"response to the show cause order stated that she used generative artificial\" intelligence to research authorities and failed to verify them. The writ petition was denied and $1,750 imposed under Mo. R. Civ. P. 55.03. Two framing points matter for any eventual record. The sanction runs as attorney’s fees payable to the opposing private parties, not as a fine to the court, so the tracker’s amount is right but its character is not. And the panel expressly disclaims sanctioning the use of AI itself: \"The use of AI is only incidental to the sanction.\" That is a framing statement rather than a displacement, since the order’s own mechanism for the fabrication is unverified generative-AI research. The judge named in the caption is the respondent by writ convention and used no AI. Held on gate 2 alone: courts.mo.gov returned HTTP 500 and then an empty body for its opinion index across repeated attempts, and CourtListener carries no opinion for this matter.", "next_verification_step": "Retrieve the order from the Missouri Court of Appeals, Eastern District, through courts.mo.gov when its opinion index is serving, and confirm the panel, the obligor and that the $1,750 runs to the underlying petitioners rather than to the court.", "best_available_sources": "", "last_checked_date": "2026-09-22" } ] }, "global": { "label": "Global candidates", "description": "Non-US and global candidate matters requiring translation, event-grain decisions, or source hardening.", "records": [ { "candidate_id": "AIEL-GLOB-001", "candidate_title": "Robodebt automated welfare-debt income-averaging scheme", "country": "Australia", "region": "APAC / Commonwealth", "event_grain": "systemic_error_program", "candidate_matter": "Prygodicz v Commonwealth of Australia (No 2) / Royal Commission into the Robodebt Scheme", "authority_type": "court; royal commission; agency", "legal_basis": "social security law; administrative law; public administration", "reason_for_review": "Strong global candidate with official anchors, but should be modeled as a systemic program rather than an individualized error event. Needs original Federal Court decision URL and final/current settlement posture captured.", "next_verification_step": "Retrieve FCA 634 full judgment from Federal Court/AustLII/Jade; reconcile 2021 settlement, Royal Commission findings, and later class-action settlement developments; decide whether to create one systemic row or multiple rows by legal milestone.", "best_available_sources": "https://fedcourt.gov.au/digital-law-library/judges-speeches/justice-kyrou/kyrou-j-20240318; https://robodebt.royalcommission.gov.au/publications/report; https://hrlc.org.au/case-summaries/2021-9-30-the-federal-court-approves-a-112-million-settlement-for-the-failures-of-the-robodebt-system/", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-002", "candidate_title": "SyRI welfare-fraud risk-indication system", "country": "Netherlands", "region": "EU/EEA", "event_grain": "legality_challenge_without_specific_error", "candidate_matter": "NJCM c.s. v Staat der Nederlanden, ECLI:NL:RBDHA:2020:865", "authority_type": "court", "legal_basis": "ECHR Article 8; privacy; data protection; administrative law", "reason_for_review": "Strong landmark case, but court record is a legality/privacy challenge to an opaque risk system rather than a single proved erroneous output. Candidate should be tagged systemic/legality unless later evidence identifies specific adverse decisions.", "next_verification_step": "Translate and review official judgment; capture original Dutch matter name; decide whether to include as legality-challenge row or keep in separate systemic-program table.", "best_available_sources": "https://uitspraken.rechtspraak.nl/details?id=ECLI%3ANL%3ARBDHA%3A2020%3A865; https://privacyinternational.org/news-analysis/3363/syri-case-landmark-ruling-benefits-claimants-around-world", "source_language": "Dutch", "translation_status": "machine translated", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-003", "candidate_title": "Uber/Ola automated driver management and robo-firing litigation", "country": "Netherlands", "region": "EU/EEA", "event_grain": "legality_challenge_without_specific_error", "candidate_matter": "Amsterdam Court of Appeal Uber/Ola algorithmic decision-making cases", "authority_type": "court", "legal_basis": "GDPR Article 22; GDPR Articles 13-15; worker data rights", "reason_for_review": "Useful GDPR/platform-work candidate, but needs exact ECLI records and row-splitting between Uber deactivation, Ola deductions, pay/pricing, and disclosure issues.", "next_verification_step": "Find official Amsterdam Court of Appeal ECLI judgments; split by company and automated decision type; verify which decisions had legal/significant effects and which remedies were ordered.", "best_available_sources": "https://fountaincourt.uk/2023/04/amsterdam-court-upholds-appeal-in-algorithmic-decision-making-test-case-drivers-v-uber-and-ola/; https://ekker.legal/wp-content/uploads/2024/09/Historic-digital-rights-win-for-WIE-and-the-ADCU-over-Uber-and-Ola-at-Amsterdam-Court-of-Appeal.pdf", "source_language": "Dutch / English summaries", "translation_status": "needs translation", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-004", "candidate_title": "Deliveroo Frank rider-ranking algorithm", "country": "Italy", "region": "EU/EEA", "event_grain": "systemic_error_program", "candidate_matter": "Filcams CGIL Bologna and others v Deliveroo Italia SRL / Bologna Labour Court", "authority_type": "labor tribunal / court", "legal_basis": "labor law; anti-discrimination; right to strike", "reason_for_review": "Strong algorithmic-management candidate, but primary Italian judgment should be located before inclusion. Likely systemic row because the algorithmic ranking rule affected access to shifts rather than one named claimant event.", "next_verification_step": "Locate official or union-hosted Italian judgment PDF; preserve original title; verify remedy amount and whether ruling was appealed or superseded by platform changes.", "best_available_sources": "https://techcrunch.com/2021/01/04/italian-court-rules-against-discriminatory-deliveroo-rider-ranking-algorithm/; https://apps.eurofound.europa.eu/platformeconomydb/italian-court-rules-that-the-deliveroo-algorithm-discriminates-workers-106121; https://cliffordchance.com/insights/resources/blogs/talking-tech/en/articles/2021/06/the-italian-courts-lead-the-way-on-explainable-ai.html", "source_language": "Italian / English summaries", "translation_status": "needs translation", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-005", "candidate_title": "CNAF French family-benefits risk-scoring algorithm challenge", "country": "France", "region": "EU/EEA", "event_grain": "legality_challenge_without_specific_error", "candidate_matter": "La Quadrature du Net et al. v CNAF / Conseil d'État challenge", "authority_type": "administrative court; NGO complaint", "legal_basis": "GDPR Article 22; non-discrimination; privacy; AI Act social-scoring relevance", "reason_for_review": "Important ongoing EU public-benefits risk-scoring candidate, but appears at complaint/challenge stage and needs Conseil d'État docket or filings plus any ombuds or source-code materials.", "next_verification_step": "Find Conseil d'État docket/filing if public; archive complaint; track hearing and decision; decide whether to include only after procedural order or merits ruling.", "best_available_sources": "https://laquadrature.net/en/2024/10/16/french-family-welfare-scoring-algorithm-challenged-in-court-by-15-organisations/; https://amnesty.org/en/latest/news/2024/10/france-discriminatory-algorithm-used-by-the-social-security-agency-must-be-stopped/; https://edri.org/our-work/cnafs-discriminatory-scoring-algorithm-10-new-organisations-join-the-case-before-the-conseil-detat-in-france/", "source_language": "French / English summaries", "translation_status": "needs translation", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-006", "candidate_title": "Alvi Choudhury retrospective facial-recognition mismatch arrest", "country": "United Kingdom", "region": "UK", "event_grain": "individualized_error_event", "candidate_matter": "Potential damages claim against Thames Valley Police", "authority_type": "claim / pre-action; police matter", "legal_basis": "false imprisonment; data protection; ECHR Article 8; equality law", "reason_for_review": "Strong individualized error candidate, but public materials currently appear claimant-side and media-based; no court claim number or filed public complaint found yet.", "next_verification_step": "Monitor for issued claim, pre-action letter, settlement, or police complaint outcome; obtain court filing if claim is issued.", "best_available_sources": "https://doughtystreet.co.uk/news/automated-facial-recognition-software-innocent-man-arrested; https://dpp-law.com/blog/wrongful-arrest-by-algorithm-the-alvi-choudhury-case-and-your-rights/; https://bbc.com/news/articles/cd9gkj3lqpgo", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-007", "candidate_title": "Metropolitan Police live facial-recognition policy challenge involving false match", "country": "United Kingdom", "region": "UK", "event_grain": "legality_challenge_without_specific_error", "candidate_matter": "High Court challenge by Shaun Thompson and Silkie Carlo / Big Brother Watch context", "authority_type": "court", "legal_basis": "ECHR Article 8; privacy; freedom of expression/assembly; equality concerns", "reason_for_review": "Challenge reportedly lost, and the false match was part of the factual context rather than a successful error claim. Candidate should remain review/exclusion until judgment is reviewed.", "next_verification_step": "Fetch 74-page High Court judgment; determine whether false match should be separate event row or exclusion note due unsuccessful policy challenge.", "best_available_sources": "https://bbc.co.uk/news/articles/cq59x4vv954o", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-008", "candidate_title": "Dutch childcare-benefits algorithm / toeslagenaffaire risk profiling", "country": "Netherlands", "region": "EU/EEA", "event_grain": "systemic_error_program", "candidate_matter": "Dutch childcare allowance scandal / tax authority investigations", "authority_type": "parliamentary inquiry; data protection authority; compensation scheme", "legal_basis": "anti-discrimination; administrative law; data protection; public benefits", "reason_for_review": "Major global example, but the AI/model role is intertwined with broader administrative and political misconduct. Needs official Dutch DPA/parliamentary records and careful event-grain definition.", "next_verification_step": "Find Dutch Data Protection Authority report, parliamentary inquiry records, and compensation scheme materials; decide whether to include as algorithmic risk-profiling row or broader exclusion/adjacent row.", "best_available_sources": "https://amnesty.org/en/latest/news/2021/10/xenophobic-machines-dutch-child-benefit-scandal/; https://taxadmin.ai/taxadminai-toeslagenaffaire/", "source_language": "Dutch / English summaries", "translation_status": "needs translation", "last_checked_date": "2026-04-22" }, { "candidate_id": "AIEL-GLOB-009", "candidate_title": "Israel National Labor Court sanctions attorney 2,500 ILS for AI-fabricated citations", "country": "Israel", "region": "MENA", "event_grain": "single_event", "candidate_matter": "Kibbutz Beit Nir v. Ronen Marciano, Case בר\"ע 34038-02-26 (National Labor Court)", "authority_type": "court", "legal_basis": "professional conduct; court sanctions authority", "reason_for_review": "Primary source in Hebrew; court finding of AI-generated fabricated citations confirmed by pdftotext extraction. Gate 1 passes (unnamed-court-found). Needs translation for full record. Decision is documented as the first instance of personal costs imposed on an attorney for AI-generated erroneous citations in the National Labor Court.", "next_verification_step": "Obtain Hebrew-to-English translation of the decision text for accurate error_description and canonical quote.", "best_available_sources": "", "source_language": "Hebrew", "translation_status": "needed", "last_checked_date": "2026-06-02" }, { "candidate_id": "AIEL-GLOB-010", "candidate_title": "Spain TSJ Canarias sanctions attorney €420 for AI-hallucinated citations; first reported Spanish appellate AI sanction", "country": "Spain", "region": "EU/EEA", "event_grain": "single_event", "candidate_matter": "Pieza Sanciones Mala Fe Procesal, Recurso de apelación Nº 0000140/2025-00 (TSJ de Canarias, Sala Civil y Penal)", "authority_type": "court", "legal_basis": "Art. 247.3 LEC / Art. 554.2 LOPJ (procedural bad faith / temeridad procesal)", "reason_for_review": "Gate 1 passes (unnamed paid-subscription generative AI tool, attorney-admitted). Primary source in Spanish. Needs translation for full record. Also includes mandatory referral to the Colegio de Abogados de Santa Cruz de Tenerife (bar disciplinary). Reported as first Spanish appellate AI sanctions decision.", "next_verification_step": "Obtain translation of key passages for error_description and canonical quote; confirm this is the first reported Spanish case.", "best_available_sources": "", "source_language": "Spanish", "translation_status": "needed", "last_checked_date": "2026-06-02" }, { "candidate_id": "AIEL-GLOB-011", "candidate_title": "Colombian Supreme Court (Civil) fines attorney 15 SMLMV for AI-generated fabricated citations", "country": "Colombia", "region": "LATAM", "event_grain": "single_event", "candidate_matter": "Auto AC739-2026, Rad. 11001-02-03-000-2025-05324-00 (Corte Suprema de Justicia, Sala de Casación Civil)", "authority_type": "court", "legal_basis": "Arts. 79-81 Código General del Proceso (procedural recklessness / temeridad procesal)", "reason_for_review": "Gate 1 passes (attorney expressly admitted AI-program use generated the inaccurate citations). Primary source in Spanish. Fine is 15 SMLMV (~USD 5,300 at 2026 rates) — a real and substantial penalty, not a nominal placeholder. Corte Suprema de Justicia is Colombia's highest court for civil matters (Sala de Casación Civil). Magistrate ponente: Martha Patricia Guzmán Álvarez.", "next_verification_step": "Obtain translation of key passages for corpus record; verify 2026 SMLMV rate for USD equivalent.", "best_available_sources": "", "source_language": "Spanish", "translation_status": "needed", "last_checked_date": "2026-06-02" }, { "candidate_id": "AIEL-GLOB-012", "candidate_title": "Australian federal court refers solicitor to LPCC after Google AI Overviews generated false immigration citations", "country": "Australia", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Pasuengos v Minister for Immigration and Citizenship (No 2) [2026] FedCFamC2G 96 (Federal Circuit and Family Court of Australia, Division 2)", "authority_type": "court", "legal_basis": "Court costs jurisdiction; referral to Legal Profession Conduct Commissioner (SA)", "reason_for_review": "Gate 1 passes on admission prong: counsel expressly admitted on the record that citation research was generated by a Google Search AI summary (Google AI Overviews), which produced incorrect case citations that were not verified. Note: tool is Google AI Overviews (a generative-AI-powered search feature), not a standalone LLM drafting tool — flag for human determination of whether this falls within corpus scope, consistent with the decision to exclude the Doiban (Oregon) matter on similar grounds. Costs: AUD $3,125 personal undertaking by the applicant's representative plus LPCC referral. Primary source in English.", "next_verification_step": "Decide whether Google AI Overviews citations qualify under corpus gate-1 scope (cf. Doiban decision to exclude); if yes, promote to included.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-06-02" }, { "candidate_id": "AIEL-GLOB-013", "candidate_title": "German court holds Google directly liable for false AI Overviews statements about two Munich publishers", "country": "Germany", "region": "EU/EEA", "event_grain": "single_event", "candidate_matter": "LG München I, Endurteil v. 28.05.2026 - 26 O 869/26 (einstweilige Verfügung; two publishers v. Google)", "authority_type": "court", "legal_basis": "§§ 1004, 823 Abs. 1 BGB i.V.m. Art. 2 Abs. 1, Art. 19 Abs. 3 GG (corporate personality right); DSA Art. 6(1) hosting privilege held inapplicable because AI Overview statements are the search engine operator's own; EU AI Act (KI-VO) held inapplicable", "reason_for_review": "Party names redacted in published certified copy; decision text currently hosted only by news outlet (the-decoder.com), not yet in dejure/openjur/gesetze-bayern; injunction partially rejected (item 3) and judgment appealable — final posture open", "next_verification_step": "Locate decision in an official German database (gesetze-bayern.de, openjur.de) or LG München I press release; identify plaintiff publishers and confirm whether Google appeals to OLG München", "best_available_sources": "https://the-decoder.com/wp-content/uploads/2026/06/26_O_869_26_begl_Abschrift_Urteil_v_28_05_2026_Geschwarzt_Geschwarzt_Geschwarzt.pdf; https://the-decoder.com/landmark-german-ruling-declares-googles-ai-overviews-are-googles-own-words-and-makes-it-liable-for-false-answers/", "source_language": "German", "translation_status": "partial", "last_checked_date": "2026-06-10" }, { "candidate_id": "AIEL-GLOB-014", "candidate_title": "Supreme Court of India sets aside NCLT and NCLAT orders that rested on AI-generated fake citations in the tribunal’s own judgment", "country": "India", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd. & Anr., Civil Appeal No. 11950 of 2025, 2026 INSC 668 (Supreme Court of India)", "authority_type": "court", "legal_basis": "Insolvency and Bankruptcy Code s. 7 (corporate insolvency resolution); doctrine that a decision resting on fake or hallucinated material is no decision at all; Supreme Court direction to the Bar Council of India to constitute a committee on fake and hallucinated citations", "reason_for_review": "Non-US matter requiring jurisdiction-specific framing, and the corpus’s first adjudicator-side record. Gate 1 passes on the Court’s own express finding (\"the Tribunal relied on non-existent, fake and hallucinated material, generated through Artificial Intelligence (AI)\"), and para 15 treats the non-existence of the six cited judgments as not in dispute. No AI product is named. The distinguishing fact is that the fabricated citations came from the adjudicating authority’s own research, not from counsel: Respondent No. 1’s affidavit confirmed its counsel never cited those judgments at the bar. The linked PDF is the official Supreme Court e-copy (digitally signed by the registry, 2026-07-02) but is rehosted by a third party; an official database URL is still needed. The s. 7 application was restored for fresh decision, so the underlying matter remains live.", "next_verification_step": "Locate the judgment on an official Supreme Court of India source (sci.gov.in judgment portal or the 2026 INSC 668 neutral-citation entry) to replace the rehosted copy, then confirm the outcome of the restored application RCP (IB) 6/MB/2023 and whether the Bar Council of India committee issues guidance.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-07-24" }, { "candidate_id": "AIEL-GLOB-015", "candidate_title": "Ontario law-society tribunal awards full costs against a suspended lawyer who filed Grok-generated fake citations in his own regulatory case", "country": "Canada", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Mazaheri v Law Society of Ontario, 2026 ONLSTH 112, Tribunal File No. 25H-139 (Law Society Tribunal, Hearing Division, Ontario)", "authority_type": "tribunal", "legal_basis": "Law Society Tribunal costs jurisdiction (compensatory, not punitive); the panel expressly held it had no authority to make a professional-misconduct finding about the AI use because the Law Society had brought no conduct application on that issue", "reason_for_review": "Gate 1 passes clearly: the licensee admitted in writing that errors in his factum and supplementary factum “arose for my over-reliance on generative artificial intelligence tools (in particular Grok)”, and the panel made its own findings, including that “the tool the applicant used was hallucinating”, under section headings referring to irresponsible use of artificial intelligence. Held in review rather than admitted because of two sourcing and posture problems. Sourcing: the only machine-verified original is the tribunal’s own one-paragraph order page, which carries the operative costs order but not the 34 pages of reasons containing the AI findings; the full reasons exist only on CanLII, which serves HTTP 403 to all automated requests by robots policy, so no citable link to the reasons has been verified. Posture: the costs order was STAYED on 2026-07-23 pending an appeal filed 2026-07-13 under Appeal File 26A-020, so the CAD 31,150 is not a settled outcome. Held on a real gap, not on jurisdiction: under the rule settled in INTENT.md on 2026-08-19, `global` is `review` for non-US matters and a record whose only remaining gap is that it is not American has no remaining gap. Criterion 3 is also unsettled here because the costs order was stayed on 2026-07-23 pending appeal (Appeal File 26A-020), so this candidate is held on both sourcing and posture.", "next_verification_step": "Confirm the CanLII address for 2026 ONLSTH 112 in a browser (canlii.ca/t/klgxt), or obtain a CanLII API key so Canadian records can be verified without hitting bot walls, then record the reasons as the primary source; separately track Appeal File 26A-020 and re-verify once the stay is resolved.", "best_available_sources": "https://lawsocietytribunal.ca/document-html/mazaheri-2026-06-12/; https://lawsocietytribunal.ca/licensee/shahryar-mazaheri/; https://lawsocietytribunal.ca/document-html/mazaheri-2026-07-23/", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-GLOB-017", "candidate_title": "Quebec Court of Appeal orders self-represented appellant to pay costs over conceded ChatGPT authorities", "country": "Canada", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Droit de la famille — 261234, 2026 QCCA 1092, No. 500-09-700419-252 (Court of Appeal of Quebec, Montreal, 11 August 2026)", "authority_type": "court", "legal_basis": "Appellate costs jurisdiction; abusive conduct in the conduct of an appeal", "reason_for_review": "Gate 1 passes: the self-represented appellant conceded that the authorities he cited are the product of the artificial intelligence ChatGPT, and the court held that a litigant cannot assume the result obtained is free from hallucinations produced by the engine used. Gate 2 currently fails for sourcing only: the judgment is published on CanLII and SOQUIJ, both of which block automated retrieval (CanLII returns HTTP 403 behind a DataDome CAPTCHA; SOQUIJ exposes no stable per-judgment deep link and its citizen portal returned 404 for this neutral citation). No original court-hosted URL was reachable, so no link is recorded. Held on a real gap, not on jurisdiction: under the rule settled in INTENT.md on 2026-08-19, `global` is `review` for non-US matters and a record whose only remaining gap is that it is not American has no remaining gap.", "next_verification_step": "Obtain a free CanLII API key (api.canlii.org, which returns HTTP 401 unauthenticated) and retrieve 2026 QCCA 1092 through it; this would also unblock every future Canadian record. The court declined the respondent's $5,000 claim and instead condemned A. P. to pay $1,000 in legal costs within 30 days.", "best_available_sources": "", "source_language": "French", "translation_status": "not needed", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-GLOB-018", "candidate_title": "Gauteng full bench refers acting judge's false citations to Legal Practice Council without finding AI use", "country": "South Africa", "region": "Africa", "event_grain": "single_event", "candidate_matter": "FJL v TGO, Case No. 25-220239 (High Court of South Africa, Gauteng Division, Johannesburg, 31 July 2026)", "authority_type": "court", "legal_basis": "Superior Courts Act section 18(3)/(4) appeal; Judicial Service Commission Act referral jurisdiction; Legal Practice Act", "reason_for_review": "Admissible in principle as of 2026-08-19; **blocked on gate 2 sourcing only**. Under the undetermined-attribution rule in INTENT.md, criterion 1 is satisfied: the second judgment records at paragraph 37 that the material bears the hallmarks of AI hallucination while making no finding on whether AI was used, and the majority declined at paragraph 18 to enter the lists about AI. That is the tribunal's own AI language in connection with the defect, and no competing non-AI mechanism is established. The acting judge's judgment below contained discrepancies in cited judgments, including a quotation partly drawn from an unrelated case and conflicting citations for MEC for Health Eastern Cape v Mkhintha, and the matter was referred to the Legal Practice Council. Gate 2 fails on sourcing: SAFLII returns HTTP 403 to automated requests on every path and no other court-hosted copy was reachable, so South African judgments remain browser-only.", "next_verification_step": "Obtain the judgment through a human browser pass on SAFLII or another South African court-hosted source, then promote with the attribution qualifier, the `attribution-not-determined` tag and the paragraph 18 and 37 language quoted. This record is adjudicator-side, so retain the distinction between the acting judge as the AI-implicated actor and the full bench as the reviewing tribunal.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-GLOB-019", "candidate_title": "Nova Scotia Court of Appeal awards $20,000 solicitor-client costs over AI-fabricated transcripts and appeal books", "country": "Canada", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Arbuckle v. Tanner, 2026 NSCA 62 (Nova Scotia Court of Appeal, 12 August 2026)", "authority_type": "court", "legal_basis": "Appellate costs jurisdiction, Nova Scotia Civil Procedure Rule 90.40(4); abuse of process", "reason_for_review": "Gate 1 passes on the appellant's own admission: the self-represented appellant Kevin Joel Arbuckle swore an affidavit on 24 July 2026 stating that, unbeknownst to him during the assembly process, \"the AI hallucinated\" the certifications at issue, and the court records that other evidence confirms this. The matter is materially graver than a fabricated-citation case: the court found the transcripts and appeal books themselves to be non-genuine, and addressed conduct amounting to reproducing a court official's signature or fabricating a certified court record. Costs of $20,000 were awarded to the respondent on a solicitor-client basis under Rule 90.40(4). Gate 2 fails for sourcing only: no court-hosted original was reachable. The Nova Scotia judgment site decisions.courts.ns.ca answers on its root but returned HTTP 404 for every navigation and search path attempted, and CanLII remains blocked behind a DataDome CAPTCHA. Held on a real gap, not on jurisdiction: under the rule settled in INTENT.md on 2026-08-19, `global` is `review` for non-US matters and a record whose only remaining gap is that it is not American has no remaining gap.", "next_verification_step": "Obtain the free CanLII API key already flagged in the skill's open follow-up work (api.canlii.org returns HTTP 401 unauthenticated) and retrieve 2026 NSCA 62 through it; this would unblock this record and AIEL-GLOB-017 together. Failing that, a human browser pass through decisions.courts.ns.ca to establish the working item-URL pattern would be worth recording in the skill's per-forum sourcing table, since Nova Scotia is not yet mapped there.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-GLOB-020", "candidate_title": "Australian federal court fixes costs at $9,097.93 and holds that neither the litigant, ChatGPT nor OpenAI can be referred to a regulator", "country": "Australia", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Asif v Minister for Immigration and Citizenship (Federal Circuit and Family Court of Australia, Division 2, General Federal Law, 22 July 2026)", "authority_type": "court", "legal_basis": "Judicial review of a migration decision; costs jurisdiction; procedural orders as to written submissions", "reason_for_review": "Gate 1 passes as named and admitted: asked at the hearing whether he had used generative artificial intelligence to produce his submissions, the self-represented applicant \"quite candidly told the Court he had used ChatGPT.\" The court found the resulting submissions unnecessarily lengthy, repetitive and of little substantive content, and treated citation and quotation errors as indicators of probable generative-AI use. The judgment is of interest to the corpus beyond the sanction: the judge recounts interrogating ChatGPT about one of his own judgments and reproduces its admission that it fabricated the details, and holds that unrepresented applicants, ChatGPT itself and OpenAI as its owner owe no professional duty to the court and cannot be referred to a legal regulator when references are hallucinated. That is an enforcement-gap holding the corpus does not otherwise cover. Gate 2 fails for sourcing only: fcfcoa.gov.au timed out from this network on both the bare and www hosts, and AustLII returned HTTP 403 to scripted requests and is an aggregator in any event.", "next_verification_step": "Retrieve the judgment from the Federal Circuit and Family Court of Australia's own judgment database at fcfcoa.gov.au, which requires a network path that resolves that host, and capture the medium-neutral citation, which was not visible in the copy inspected. Confirm on retrieval that costs were fixed in the sum of $9,097.93 payable to the First Respondent; the tracker records the penalty as \"1 AUD\", which is wrong.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-08-19" }, { "candidate_id": "AIEL-GLOB-021", "candidate_title": "Gujarat tax officer's order relying exclusively on AI-hallucinated case law quashed; triggers statewide AI-governance instructions", "country": "India", "region": "APAC / Commonwealth", "event_grain": "single_event", "candidate_matter": "Faiz Enterprise v. State Tax Officer, Unit-67, Surat & Anr., R/Special Civil Application No. 10123 of 2026 (Gujarat High Court)", "authority_type": "court; administrative/quasi-judicial tax authority", "legal_basis": "GST Act adjudication; administrative law; natural justice", "reason_for_review": "Strong adjudicator-side matter, but the AI-using party is a State Tax Officer exercising quasi-judicial GST adjudication (a probationary officer), not a judge — the tracker's \"Judge\" tag over-claims and needs correction before this could be drafted as a judicial-AI-use record. Bucket is `global` on the non-US rule alone; sourcing is fully resolved via the court's own QR-verified copy.", "next_verification_step": "Steward review of the systemic-outcome angle: the Additional Commissioner of State Tax (Admin), Gujarat issued statewide instructions (No. CCT/Admin/instructions/2026-27/O.N:-53, 18 Aug 2026) governing AI use by adjudicating and quasi-judicial authorities, which the High Court incorporated into its order with contempt-of-court backing for violations — this may be worth its own Obligation-First anchor as an AI-governance instrument distinct from the underlying incident.", "best_available_sources": "https://gujarathc-casestatus.nic.in/gujarathc/VerifyNCQR?aid=MjAyNi1HSkhDMjQwNDg0MjEyMDI2LTItdW5kZWZpbmVk", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-09-02" }, { "candidate_id": "AIEL-GLOB-022", "candidate_title": "Tel Aviv District Court finds reliance on AI hallucinations in class-certification withdrawal, awards costs", "country": "Israel", "region": "MENA", "event_grain": "single_event", "candidate_matter": "Adibi v. AIG Israel Insurance Co. Ltd., CA (class action) 6989-05-25 (Tel Aviv-Jaffa District Court)", "authority_type": "court", "legal_basis": "Class Actions Law 5766-2006", "reason_for_review": "Gate 1 passes on the court's own repeated Hebrew-language finding (\"the reliance on AI hallucinations\") across five fabricated citations, but sourcing is currently limited to the tracker's own document server plus one law-firm blog (law.co.il) — no Nevo, net.hamishpat, or court.gov.il citation located. The tracker's \"Monetary Fine\" tag is materially wrong: the ILS 20,000 is an ordinary adverse-party costs award against the plaintiff, not a sanction on counsel — the court considered and expressly declined a personal-costs sanction against counsel, settling for a warning. No citable court-issued original could be located (checked net.hamishpat.gov.il patterns and Nevo); the only source found is a law-firm blog summary. Matter is cited here by its official case number (CA 6989-05-25, Tel Aviv-Jaffa District Court) pending a reachable original.", "next_verification_step": "Locate an official Israeli judiciary publication (net.hamishpat.gov.il or Nevo) for the underlying order; confirm final translation of the operative Hebrew passages before any promotion to `included`.", "best_available_sources": "https://law.co.il/computer-law/2026/08/29/adiby-v-aig-israel-insurance-company-ltd/", "source_language": "Hebrew", "translation_status": "partial", "last_checked_date": "2026-09-02" }, { "candidate_id": "AIEL-GLOB-023", "candidate_title": "Supreme Court of India sets aside a customs penalty order built on AI-hallucinated precedent", "country": "India", "region": "APAC / Commonwealth", "event_grain": "individualized_error_event", "candidate_matter": "Vijay Ghanshyam Gadiya v. Union of India & Anr., 2026 INSC 947, Civil Appeal arising out of SLP (Civil) No. 15605 of 2026 (Supreme Court of India, 2 September 2026)", "authority_type": "court; administrative/quasi-judicial customs authority", "legal_basis": "Section 114, Customs Act 1962; principles governing reliance on non-existent precedent in quasi-judicial adjudication", "reason_for_review": "Gate 1 is satisfied in the Court’s own voice after independent checking: it found the authorities relied on in the Order-in-Original \"are generated using Artificial Intelligence\" and that a passage \"appears to be a hallucination of AI\". Adjudicator-side, but the AI-using adjudicator is an Additional Commissioner of Customs exercising quasi-judicial power, not a judicial officer, so the tracker’s \"Judge\" tag over-claims; no judge of the High Court or the Supreme Court used AI. The Court set aside both the High Court order and the Order-in-Original imposing a penalty of Rs 425,27,99,100 and remitted the matter to a different officer of the same rank, leaving action against the author of the order to the appointing authority. The judgment goes beyond the individual matter, distinguishing legitimate assistive use of AI from reliance on fabricated material as though it were precedent, and referring to draft Regulations for Use of Artificial Intelligence in Courts, 2026. Held on gate 2 alone: no citable original was reachable. A guessed api.sci.gov.in judgment path returned HTTP 200 with a zero-byte body, digiscr.sci.gov.in does not resolve, and the eCourts judgment portal is a JavaScript search application behind tokens.", "next_verification_step": "Retrieve the signed judgment from the Supreme Court of India, either through the eCourts judgment portal in a browser or by resolving the digital-signature or QR verification route printed on the signed copy, and confirm the neutral citation 2026 INSC 947, the bench and the penalty figure before promotion. The draft Regulations for Use of Artificial Intelligence in Courts, 2026 are separately worth routing to EveryAILaw.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-09-22" }, { "candidate_id": "AIEL-GLOB-024", "candidate_title": "Australian court orders personal costs against a practitioner over eleven hallucinated authorities but declines to refer him to the regulator", "country": "Australia", "region": "APAC / Commonwealth", "event_grain": "individualized_error_event", "candidate_matter": "Dey v Minister for Immigration and Citizenship (No 2) [2026] FedCFamC2G 1799 (Federal Circuit and Family Court of Australia, Division 2, Judge Bingham, 14 September 2026)", "authority_type": "court", "legal_basis": "Costs against a practitioner personally; Rule 3.1, Legal Profession Uniform Law Australian Solicitors’ Conduct Rules 2015 (paramount duty to the court)", "reason_for_review": "Gate 1 is satisfied on admissions recorded in the reasons: the practitioner \"admits that six authorities referred to in the judicial review submissions were hallucinations and five of the\" authorities in the First Costs Submission were also hallucinations, and he did so only after the Court sought clarification. Costs were ordered in two parts, $6,700 payable by Mr Dey and $9,600 payable personally by the practitioner; the tracker reports only the second figure and its penalty column is unusable. The reason to admit this record is the contrast it draws with AIEL-GLOB-020 (Asif), which holds that a lay litigant owes no duty to the court and so cannot be referred to any regulator. Here the respondent is a practitioner who does owe that duty and referral was available, yet the Court declined it, having made no finding of unsatisfactory professional conduct or professional misconduct, treating referral as an administrative act any party could take independently, and noting that declining to refer does not require anonymising the practitioner. The Court also records a practice direction on AI use that took effect 29 May 2026 and did not apply to these proceedings. Held on gate 2 alone: the retained copy is an AustLII rendering rather than the court’s own publication, AustLII is an aggregator and bars citation, and fcfcoa.gov.au was unreachable from this run behind a Radware Bot Manager CAPTCHA.", "next_verification_step": "Retrieve the reasons from the Federal Circuit and Family Court of Australia’s own judgment database through a browser pass at fcfcoa.gov.au, confirm the neutral citation [2026] FedCFamC2G 1799 and both costs figures, and establish what the first judgment, [2026] FedCFamC2G 373, decided and whether it warrants its own record. The Court’s AI practice direction of 29 May 2026 is separately worth routing to EveryAILaw.", "best_available_sources": "", "source_language": "English", "translation_status": "not needed", "last_checked_date": "2026-09-22" } ] } } }