reportAI HallucinationsAI in CourtsAI Governance
Legal AI incident report, Q3 2026: 87 incidents, 61 in US courts, 26 regulator dispositions
1 July to 30 September 2026: 87 tracker incidents in 11 jurisdictions, USD 142,729 in stated US penalties, 26 regulator referrals or dispositions.
Edited and verified by SafeLegalAI · updated
Researched with AI assistance · sources verified by SafeLegalAI · How this was made ↓
Updated 7 October 2026: full-quarter figures. This report was first published on 3 September 2026 and covered Q3 2026 only to that date (54 incidents). With the quarter now closed, it has been recomputed for the full window 1 July to 30 September 2026 from the incident records as of 7 October 2026: 87 incidents, 61 in United States courts, 26 regulator referrals or dispositions dated in the quarter. The title, description, headline and every table below now report the full quarter. The superseded 3 September figures are preserved in the dated note at the end of this section; no partial figure has been changed silently.
As of 7 October 2026 the SafeLegalAI Global Legal AI Incident Tracker holds 87 incidents dated in the third quarter of 2026 (1 July to 30 September 2026), against 8 in the second quarter and 13 in the third quarter of 2025. Sixty-one of the 87 are in United States courts, federal and state; the other 26 are spread across Canada, Australia, India, Brazil, Israel, South Africa, the United Kingdom, the Netherlands and China.
Of the 87 rows, 80 are verified against a primary document and 7 are flagged unverified. The tracker holds 173 non-provisional rows in total as of 7 October 2026; a further 476 provisional rows, produced by the court-decisions pipeline and awaiting editorial re-reading, are excluded from every count, statistic, feed and export here, exactly as they are excluded from the public /tracker/incidents.json export. Every number below is computed from the incident records by a script saved in the run log, and the same data is exported at that JSON endpoint.
Superseded figures (3 September 2026 publication). The first version reported Q3 2026 “to 3 September” because the quarter was not over: 54 incidents (47 verified, 7 unverified), 33 in US courts, across 11 jurisdictions; 20 rows carrying a regulatory referral or disposition; USD 132,888 in stated US penalties across 14 rows; a tracker total of 122 rows. The full-quarter recompute of 7 October 2026 adds the decisions dated 3–30 September and reflects continued backfill of July and August rows: 87 incidents, 61 in US courts, 26 regulator dispositions dated in the quarter, USD 142,729.25 across 19 US rows. The jump is new decisions and backfill together, not a rise in the rate of misconduct.
How many incidents were recorded in Q3 2026?
| Period | All rows | Verified | Unverified |
|---|---|---|---|
| Q3 2026 (1 Jul to 30 Sep) | 87 | 80 | 7 |
| Q2 2026 (1 Apr to 30 Jun) | 8 | 8 | 0 |
| Q3 2025 (1 Jul to 30 Sep) | 13 | 13 | 0 |
| Tracker total (non-provisional), 22 Jun 2023 to 7 Oct 2026 | 173 | 165 | 8 |
By month: 27 rows dated July, 42 dated August, 18 dated September. The quarter-on-quarter change is +79 rows and the year-on-year change is +74 rows, and neither figure should be read as a growth rate. The second quarter was never swept at the intensity applied to July, August and September; many Q3 rows were added by backfill passes after the quarter opened. A fairer baseline is Damien Charlotin’s AI Hallucination Cases database, a larger global count against which the tracker’s 87 is a selection, not a census.
What the tracker adds is the layer after the courtroom: 51 of its 173 non-provisional rows carry a recorded regulatory or disciplinary outcome, and 25 of those are third-quarter rows. Twenty-six regulatory dispositions or referrals carry a date inside the quarter.
Where did they happen?
| Jurisdiction | Q3 2026 | Q2 2026 | Q3 2025 | All rows |
|---|---|---|---|---|
| US federal | 33 | 2 | 2 | 53 |
| US state | 28 | 3 | 4 | 54 |
| Canada | 5 | 0 | 1 | 11 |
| Australia | 4 | 0 | 1 | 7 |
| India | 4 | 0 | 1 | 8 |
| United Kingdom | 4 | 3 | 3 | 19 |
| Brazil | 3 | 0 | 1 | 4 |
| Israel | 2 | 0 | 0 | 3 |
| South Africa | 2 | 0 | 0 | 5 |
| EU (Netherlands) | 1 | 0 | 0 | 1 |
| Other (China) | 1 | 0 | 0 | 2 |
Eleven jurisdictions appear in the quarter; fifteen appear in the tracker as a whole. The US share is 70 percent of the quarter (61 of 87) and 62 percent of all non-provisional rows (107 of 173). The UK, the tracker’s deepest non-US jurisdiction with 19 rows overall, contributed 4 this quarter: the Crown Prosecution Service’s two non-existent extradition authorities in Tobosaru v Court of Law Craiova; a litigant in person’s doctored documents in Lodhia v Twelve Trees; a litigant in person’s ChatGPT citations in Hancox v Sutherland; and the Solicitors Disciplinary Tribunal’s strike-off in SRA v Abhishek Kumar, the first time the SDT litigated a lawyer’s use of generative AI.
Two entries are firsts for the tracker rather than for the world. The Dutch bar disciplinary council decision in ‘s-Hertogenbosch is the first EU row, and the Yichang district court admonishment is the first Chinese row, recorded from a court WeChat report and flagged unverified because the decision itself has not been located.
Who misused the AI, and with what?
| Actor | Q3 2026 rows |
|---|---|
| Lawyer | 54 |
| Litigant in person | 23 |
| Other (adjudicator or official) | 5 |
| Not coded | 4 |
| Judge | 1 |
The actor field was added to the schema on 3 September 2026, so earlier-quarter rows are mostly uncoded (66 of the 173 non-provisional rows carry no actor) and no reliable cross-quarter comparison is possible yet. Within the quarter, lawyers account for 54 of the 87 rows and litigants in person for 23. The one row coded judge and the five coded other are the pattern the actor field first makes countable: an acting judge in Johannesburg, a municipal disciplinary-hearing chairperson whose ruling reached the Labour Court in Durban, an income-tax appellate officer in Jaipur, a state tax officer in Surat and other adjudicators in India, each found to have rested a decision on authorities that do not exist.
| AI tool named in the record | Q3 2026 | All rows |
|---|---|---|
| Not named in the record | 62 | 104 |
| Generative AI, product not identified | 9 | 22 |
| ChatGPT | 8 | 27 |
| More than one product named | 3 | 5 |
| Claude | 1 | 2 |
| Westlaw AI tools | 1 | 2 |
| Strongsuit | 1 | 1 |
| Harvey | 1 | 1 |
| AI admitted but tool unnamed | 1 | 1 |
In 62 of 87 rows (71 percent) the judgment or order does not identify the product; a further 9 rows record only that generative AI was used without a name. The tracker never infers one. Where a tool is named it is usually because counsel admitted it: ChatGPT in eight rows, Claude with other platforms in Kleyman, Westlaw’s AI tools in Rosslyn2016, where counsel specifically denied using ChatGPT. Hill v Foundation Media is the first tracker row in which a court made counsel name a legal-specific platform, Harvey, as the tool behind the suspect quotations. Across all 173 non-provisional rows, ChatGPT is named in 27 and every other product in five or fewer.
What did courts do?
| Outcome | Q3 2026 | Q2 2026 | All rows |
|---|---|---|---|
| Warning | 23 | 2 | 33 |
| Sanctions | 20 | 1 | 41 |
| Other (order set aside, admonishment, self-report) | 13 | 1 | 28 |
| Fine | 9 | 0 | 12 |
| Referral | 7 | 1 | 20 |
| Costs order | 5 | 0 | 12 |
| Pending | 5 | 0 | 6 |
| Dismissal | 2 | 2 | 16 |
| Suspension | 2 | 1 | 3 |
| Strike-off | 1 | 0 | 1 |
The commonest recorded outcome of the quarter is a warning (23 rows), ahead of sanctions (20). The single strike-off is SRA v Abhishek Kumar: the Solicitors Disciplinary Tribunal struck a Registered Foreign Lawyer off the register after he filed AI-generated false citations in his own defence, then used AI again in his correction. Monetary penalties are summed only over rows where the order states a figure. Twenty-five of the 87 rows carry a figure.
| Currency | Rows with a figure | Sum | Largest |
|---|---|---|---|
| USD | 19 | 142,729.25 | 46,511 (Kleyman Law Group v Kaloidis) |
| CAD | 3 | 31,000 | 20,000 (Arbuckle v Tanner) |
| AUD | 1 | 9,097.93 | 9,097.93 (Asif v Minister for Immigration) |
| ILS | 1 | 6,000 | 6,000 (Haifa Labor Court) |
| EUR | 1 | 1,250 | 1,250 (Raad van Discipline ‘s-Hertogenbosch) |
The USD figures mix sanctions payable to a court fund, fee-shifting to the opposing party and reimbursements. The largest, USD 46,511 in Kleyman, is a USD 10,000 sanction plus USD 36,511.35 in fees and costs. The USD 14,500 in Full Standard Properties v SJ 6070 is a reimbursement counsel paid before the show-cause hearing, after which the order was discharged. The median stated USD penalty in the quarter is USD 2,341.25.
Five rows carry the quarter’s weight, and each is traced to its primary document:
- Kleyman Law Group v Kaloidis (Supreme Court of New York, Kings County, 4 August). A lawyer suing a former client’s estate filed 19 submissions citing 23 fictitious decisions and misstating at least 83 real ones, drafted with Claude and other tools. Sanction and fees of USD 46,511; referral to the Grievance Committee.
- SRA v Abhishek Kumar (Solicitors Disciplinary Tribunal, 25 August). The Tribunal struck a Registered Foreign Lawyer off the register for AI-generated false citations and misleading quotations in his defence to SRA proceedings, with further AI-generated errors in his correction. It is the first SDT case to litigate a lawyer’s use of generative AI, and the quarter’s only strike-off.
- Scott v Illinois Human Rights Commission (Appellate Court of Illinois, 28 July). Ten false items in two briefs, attributed by counsel to a ChatGPT subscription. A USD 15,000 fine set at USD 1,500 per misstatement, with the court stating that fines must rise until they deter; the clerk was directed to send the opinion to the Attorney Registration and Disciplinary Commission.
- FJL v TGO (High Court of South Africa, Gauteng Division, full court, 31 July). At least eleven fictitious or garbled citations in an acting judge’s first-instance judgment. Opperman J forwarded the judgment to the Legal Practice Council, reasoning that acting judges remain practising advocates within its jurisdiction.
- Cervantes v Bianco (Supreme Court of California, 24 September). In a ballot-seizure writ case, the sheriff’s law firm misquoted the court’s own precedent, Common Cause v Board of Supervisors, to challenge standing. In footnote 7 the court disapproved the improper use of AI tools and referred the matter to the State Bar of California to determine whether discipline is warranted.
What did regulators do?
Twenty-five third-quarter rows carry a regulatory outcome, and 26 referrals or dispositions carry a date inside the quarter. Most are referrals or notifications with no disposition yet on record; four are verified dispositions, and three are unverified.
| Row | Body | What the record says |
|---|---|---|
| SRA v Abhishek Kumar | Solicitors Disciplinary Tribunal / SRA | Struck off the Register of Foreign Lawyers for AI-generated false citations (25 Aug); no order as to costs — a verified disposition |
| In re Brian E. Mitchell | USPTO Office of Enrollment and Discipline | Public reprimand by settlement, executed 20 and 21 July, final order 27 July, published 27 August — a verified disposition |
| Raad van Discipline ‘s-Hertogenbosch 2026:93 | Dutch bar disciplinary council | Formal reprimand for breaches of expertise and integrity; EUR 1,250 costs — a verified disposition |
| Lee v Law Society of Ontario | Law Society Tribunal (Ontario) | Six-month suspension and CAD 10,000 costs for an AI-fabricated factum (16 Jul) — now verified |
| In re Faisal Ahmed | Minnesota Supreme Court / OLPR | 30-day suspension by stipulation, reported 21 August; order not located, unverified |
| Cervantes v Bianco, Southland Homes v Lam, Varma v Bank of New York Mellon | State Bar of California | Referred or served by the court in each; no State Bar disposition yet |
| Douglas v Deutsche Bank | DC Office of Disciplinary Counsel | Brief struck; referred for any investigation it deems appropriate; no disposition |
| Kleyman Law Group v Kaloidis | Attorney Grievance Committee (New York) | Referred by the court; no disposition |
| Scott v Illinois HRC | Illinois ARDC | Opinion sent to the ARDC; no disposition |
| FJL v TGO | Legal Practice Council (South Africa) | Acting judge referred; no disposition |
| Lodhia v Twelve Trees | Bar Standards Board | Direct-access counsel referred; no disposition |
| Akerlund v Atlas Air, Parnell v Florida DOC | Eleventh Circuit Committee on Lawyer Qualifications and Conduct | Counsel referred in each; no disposition |
| JMOR Properties v Artist Alley, Capital Standard v US Bank | The Florida Bar | Counsel referred in each; no disposition |
| Forney v Township of Cary | North Carolina State Bar | Counsel referred; no disposition |
| NM Supreme Court (Aarons) | New Mexico Disciplinary Board | Referred per press report; unverified |
| Barteca v Tacobarn, Tov Realty v Suarez | Connecticut disciplinary authority / Statewide Grievance Committee | Referral in one; courtesy copy of the order in the other |
| Cartagena v Dixon Blackburn | SDNY Grievance Committee | Added to an existing referral; unverified |
| TJPR 0035599-44, TJSP 1032884-04, TRT20 Sergipe | Ordem dos Advogados do Brasil (state sections and Federal Council) | Court notices sent to the bar; no disposition; the TRT20 row is unverified |
The gap between referral and disposition is the tracker’s standing finding. In the whole record, 51 non-provisional rows carry a regulatory outcome and most are still an open referral. The four verified dispositions in the quarter came from the Solicitors Disciplinary Tribunal, the Law Society of Ontario Tribunal, a Dutch council and the US patent office — the first time an SRA-prosecuted strike-off appears on a quarter’s record. As of 7 October 2026 the State Bar of California holds the most third-quarter referrals (three) and has published no disposition on any of them.
What changed in the rules?
The Regulation Tracker holds 522 documents as of 7 October 2026, up from 84 at the 3 September publication. The bulk of that increase is a court-orders corpus added across the quarter: 102 documents carry a current-version date inside Q3 2026, of which the large majority are United States federal and state judges’ standing orders and local AI-certification rules that were already in force and were catalogued, with a catalogue date, during the quarter rather than issued in it. Read as a measure of new rule-making, that 102 overstates the quarter; read as what the tracker now covers, it is the shift from a handful of named instruments to a near-complete map of US court AI orders.
Seven documents are substantive new or revised rules of general application dated in the quarter:
| Document | Jurisdiction | New or revised | Requirements coded |
|---|---|---|---|
| Misuse of AI: warning notice (SRA, 17 Aug) | UK | New | Verification, confidentiality, supervision, competence |
| Practice Direction HC 142 (High Court of Ireland, signed 29 Jul, in operation 1 Sep) | Ireland | New | Verification, certification, disclosure, prohibition |
| Regulation (EU) 2026/1744, Digital Omnibus on AI (OJ 24 Jul) | EU | New | Risk classification |
| EU AI Act, as amended (27 Jul) | EU | Revised | Risk classification, prohibition, disclosure, competence |
| Article 50 transparency guidelines (Commission, final 20 Jul) | EU | Revised | Disclosure |
| Code of Practice on AI-generated content (Commission opinion 8 Jul, AI Board endorsement 9 Jul) | EU | Revised | Disclosure |
| Revised Ethics Guidelines on Generative AI (De Rebus, 1 Jul) | South Africa | New, draft | Verification, competence, confidentiality |
Six of the seven are in force; the South African draft has not been adopted by any regulator. The two documents that bind practitioners directly, the SRA notice and HC 142, both require verification of AI-assisted citations; HC 142 adds a declaration in affidavits and witness statements. The EU items move dates rather than duties: Annex III high-risk obligations now apply from 2 December 2027 and Annex I from 2 August 2028, while Article 50 transparency applied from 2 August 2026 as scheduled. Across all 102 Q3-dated documents the commonest coded requirements are verification (54), disclosure (54) and certification (49), reflecting the US court-order corpus; one, California SB 574, is a statute enacted 30 September.
Method and caveats
An incident is a court, tribunal or disciplinary decision in which a party, lawyer, adjudicator or official put AI-generated material before a decision-maker and the decision records it: fabricated citations, invented quotations, doctored documents, or a ruling itself built on non-existent authority. The row date is the decision date. Rows are marked verified only when the primary document has been read; 7 of the 87 quarter rows, and 8 of the 173 non-provisional rows, are unverified as of 7 October 2026 and are shown in every headline count above, alongside the verified figure. A further 476 provisional rows, produced by the court-decisions pipeline and awaiting editorial re-reading, are excluded from every count, statistic, feed and export here, exactly as they are excluded from the public JSON.
The tracker counts verified records, not every case worldwide. Damien Charlotin’s AI Hallucination Cases database is a larger global baseline; the tracker’s 87 is a verified selection, not a census. The tracker’s coverage outside the United States and United Kingdom depends on what secondary reporting surfaces, so the jurisdiction table measures where SafeLegalAI has verified rows, not where misuse is most frequent. Tool attribution is taken only from the record. Penalty sums exclude rows where the order gives no figure, and mix sanctions, fee-shifting and reimbursements. The actor field was added on 3 September 2026, so earlier-quarter rows are largely uncoded and cross-quarter actor comparisons are not reliable. The regulation count reflects a large court-orders corpus catalogued during the quarter rather than issued in it.
The dataset is exported at /tracker/incidents.json and the live figures are at /tracker/statistics. Data is licensed CC BY 4.0; cite safelegalai.com/tracker.
Three sentences journalists can quote
As of 7 October 2026, the SafeLegalAI Global Legal AI Incident Tracker records 87 verified or flagged incidents of AI misuse in legal proceedings dated between 1 July and 30 September 2026, 61 of them in United States courts.
Twenty-five of those 87 rows carry a referral to, or disposition by, a professional regulator; four regulator dispositions in the quarter are confirmed from the primary document — a strike-off by the Solicitors Disciplinary Tribunal, a six-month suspension by the Law Society of Ontario Tribunal, a USPTO public reprimand and a Dutch bar reprimand.
The largest stated penalty in the quarter is USD 46,511 in Kleyman Law Group v Kaloidis (Supreme Court of New York, Kings County, 4 August 2026), where 23 fictitious decisions were cited across 19 filings.
Sources
Primary documents for the notable rows above, each re-read for this update on 7 October 2026 unless dated otherwise:
- Kleyman Law Group, P.C. v. Kaloidis, Index No. 502644/2025 (Sup. Ct. Kings Cty. 4 Aug 2026), trial order: Charlotin archive PDF
- SRA v Abhishek Kumar, Case No. 12884-2026 (Solicitors Disciplinary Tribunal, 25 Aug 2026): SDT judgment PDF
- Cervantes v. Bianco, No. S295866 (Cal. 24 Sep 2026), footnote 7: Supreme Court of California opinion PDF
- Scott v. Illinois Human Rights Commission, 2026 IL App (1st) 251462 (28 Jul 2026): Illinois Courts opinion PDF; CourtListener
- FJL v TGO, Case No. 2025/220239 (Gauteng Division, Johannesburg, full court, 31 Jul 2026): Charlotin archive PDF
- In re Brian E. Mitchell, USPTO Proceeding No. D2026-16, final order: USPTO OED redacted PDF
- In re Rosslyn2016, LLC, No. 25-34507, ECF 312 (Bankr. S.D. Tex. 14 Jul 2026): Charlotin archive PDF
- Hill v. Foundation Media LLC, No. 1:25-cv-05947 (S.D.N.Y.), counsel’s AI-disclosure letter (Dkt. 58, 1 Oct 2026): CourtListener RECAP PDF
- Raad van Discipline ‘s-Hertogenbosch, ECLI:NL:TADRSHE:2026:93 (27 Jul 2026): tuchtrecht.overheid.nl
- Tobosaru v Court of Law Craiova, Romania [2026] EWHC 1720 (Admin): Find Case Law
- SafeLegalAI data: Incident Tracker · statistics · JSON export · Regulation Tracker · previous report: Three years of legal AI hallucinations, 2023 to 2026
Reuse this research
Quote it, cite it, forward it — CC BY 4.0 for the data and figures; the linked official documents are the record. Suggested citation and a link to this exact report: