reportAI HallucinationsAI in CourtsAI Governance
The sanctions ledger: what courts ordered over AI-fabricated authority, 2023 to 2026, priced
150 tracker rows, 45 with a stated figure: USD 371,542 across 34 US orders, median USD 5,000; 48 regulator referrals, 5 verified dispositions.
Anthropic ClaudeOpenAI ChatGPT
Edited and verified by Cognesio LLP
Researched with AI assistance · sources verified by Cognesio LLP · How this was made ↓
As of 5 September 2026 the SafeLegalAI Global Legal AI Incident Tracker holds 150 court, tribunal and regulator decisions from fifteen tracker jurisdictions (fourteen countries, the United States counted twice for its federal and state courts) in which a filing, submission or ruling rested on AI-fabricated authority, 89 of them in United States federal and state courts. This report is the ledger of what those decisions ordered: the money, the training, the referrals, the suspensions, and what the regulators who received the referrals have done since.
Forty-five of the 150 rows state a figure. In the United States, 34 orders total USD 371,542, with a median of USD 5,000 that has not moved since 2023 while the largest single matter reached USD 110,204. Forty-eight rows record a referral to a professional regulator. Five of those have a disposition that can be read in the regulator’s own document. Every number below is computed from the incident records by the script saved in the run log, and the same data is exported at /tracker/incidents.json.
Key findings
- As of 5 September 2026 the tracker holds 150 rows dated between 22 June 2023 and 3 September 2026: 6 in 2023, 11 in 2024, 40 in 2025 and 93 in the first eight months of 2026. The 2026 figure reflects a sweep of July and August decisions as much as any change in behaviour.
- Courts imposed a sanction, fine or costs order in 58 rows (39 percent), issued a warning in 27, referred conduct to a regulator as the principal outcome in 19, dismissed the case or filing in 13, and suspended a lawyer from practice in 3.
- Thirty-four US rows state a dollar figure. Their sum is USD 371,542; the median is USD 5,000 in every year from 2025 onward; 16 of the 34 are under USD 5,000 and only one exceeds USD 50,000. Remove the Oregon matter Couvrette v Wisnovsky and the 2025 total falls from USD 174,454 to USD 64,250.
- The 2026 US total to 3 September, USD 178,088 across 19 orders, already exceeds the whole of 2025. The largest 2026 order is USD 46,511 in Kleyman Law Group v Kaloidis (Supreme Court of New York, Kings County, 4 August 2026).
- In 7 rows the court ordered continuing legal education, on generative AI specifically in four of them, alone or alongside money; in 8 of the 34 US rows with a figure, the figure is wholly or mainly the opposing party’s fees rather than a penalty payable to the court.
- Forty-eight rows carry a regulator referral or disposition. Thirty-eight are referrals with nothing further on record, one records regulator action that predated the conduct, and nine record a disposition; only five of those nine can be verified from the regulator’s own published document: the Victorian Legal Services Board and Commissioner in Dayal, the four-state reciprocal chain ending in Oklahoma Bar Association v Reeves, the Law Society Tribunal of Ontario in Law Society of Ontario v Lee, the USPTO in In re Mitchell and the Dutch bar council in ‘s-Hertogenbosch 2026:93.
- Where a disposition exists, it took between 104 and 422 days to arrive: 104 days from the Alabama sanctions order in Johnson v Dunn to the first (Texas) reprimand of the same lawyer, 357 days from the Dayal referral to the practising-certificate variation, and 422 days from the first Ontario contempt ruling in Ko v Li to the Law Society Tribunal’s suspension decision.
- The court order names the AI product in 58 of 150 rows. Counting rows that may name more than one product: ChatGPT in 27, GPT-4o in 1, Claude in 5, Google products in 4, Westlaw or CoCounsel in 3 (one of them a row in which counsel denied using ChatGPT and named Westlaw’s tools), Microsoft Copilot in 2 and a practice-management or specialist tool in 7; 14 rows say only that generative AI was used or suspected, one of them listing products solely as the court’s examples. In 92 rows the order does not identify the tool, and the tracker never infers one.
Why this question
A lawyer, a general counsel or a regulator who asks “what happens when a fabricated citation reaches a court” can find plenty of headlines and several vendor-published “sanctions trackers”, most of which paraphrase Damien Charlotin’s database of court decisions. None of them tabulates the outcome side: what was actually ordered, in which currency, against whom, and whether the referral that closed the court’s file ever produced a decision by the body that licenses the lawyer. That second layer is the tracker’s reason for existing, and this report is its first full accounting.
The question matters for three practical reasons. Courts are setting per-citation rates and stating that fines must rise until they deter (Scott v Illinois Human Rights Commission). Insurers and firms are pricing the risk from press reports of the largest matters, which the median contradicts. And the regulators that received 48 referrals have published five outcomes, which is the number a practitioner should have in mind when reading any claim about “bar discipline for AI misuse”.
Method and data
The dataset is the incident collection of this site, 150 YAML records as of 5 September 2026, exported at /tracker/incidents.json. A row 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. The row date is the decision date. Rows are marked verified only when the primary document has been read; 139 of 150 are verified and 11 are flagged unverified. Unverified rows are included in counts and identified wherever they carry weight.
Fields used here: outcome (one of sanctions, fine, costs-order, referral, warning, suspension, dismissal, pending, other), monetaryPenalty and penaltyCurrency (recorded only when the order states a figure), actor (coded on 79 rows; the field was added in September 2026 and older rows have not all been re-coded), aiTool (taken only from the record), and regulatoryOutcome (body, disposition, date). Aggregations were produced by a Python script over the YAML files; the script and its output are in the run log pipeline/runs/2026-09-05-research-r1.md.
Three limits govern every figure. The tracker is a verified selection, not a census: Charlotin’s database held 2,016 entries on 4 September 2026 against the tracker’s 150, and coverage outside the United States and United Kingdom depends on what secondary reporting surfaces. Monetary figures mix penalties payable to a court, fee-shifting to an opponent and reimbursements, because the orders mix them; the report separates them where the record allows. And the outcome field records the principal outcome of the decision, so a row coded referral may also carry a fine, and a row coded sanctions may also carry a referral in its regulatoryOutcome field.
What courts ordered
Across the 150 rows the most common principal outcome is a monetary or non-monetary sanction (37 rows), followed by a warning without penalty (27), a set of outcomes coded “other” (27, mostly orders set aside, admonishments and self-reports), a referral as the main step (19), dismissal of the case or filing (13), a costs order (11), a fine (10), a suspension from practice (3) and a pending show-cause process (3).
| Principal outcome | 2023 | 2024 | 2025 | 2026 (to 3 Sep) | All rows |
|---|---|---|---|---|---|
| Sanctions | 1 | 3 | 12 | 21 | 37 |
| Warning | 1 | 2 | 3 | 21 | 27 |
| Other (set aside, admonishment, self-report) | 2 | 4 | 6 | 15 | 27 |
| Referral | 0 | 1 | 10 | 8 | 19 |
| Dismissal | 2 | 0 | 2 | 9 | 13 |
| Costs order | 0 | 1 | 3 | 7 | 11 |
| Fine | 0 | 0 | 3 | 7 | 10 |
| Suspension | 0 | 0 | 0 | 3 | 3 |
| Pending | 0 | 0 | 1 | 2 | 3 |
| Total | 6 | 11 | 40 | 93 | 150 |
The mix has changed with volume. Warnings without a penalty were 3 of 40 rows in 2025 and 21 of 93 in 2026; most of the 2026 warnings went to litigants in person in US district courts in July and August, when the tracker swept those months at full depth. Suspensions did not appear until 2026, and all three come from different routes: an appellate court’s own disciplinary power in Lnu v Blanche (Ninth Circuit, six months before that court), a law-society tribunal in Law Society of Ontario v Lee (six months, decision of 16 July 2026), and a stipulated 30-day suspension in In re Faisal Ahmed (Minnesota, reported; the order has not been located).
Reading the orders rather than the codes, the tools courts have used fall into eight kinds. A penalty payable to the court or a court fund (Mata, Wadsworth, Scott, Booker). Fee-shifting to the opposing party (Lacey, Couvrette, Rosslyn2016, Full Standard Properties). Striking the filing, dismissing the appeal or denying the relief sought (Kruse, Prososki, Ibach, Noland). Removal or disqualification as counsel (Johnson v Dunn, State v Coleman). Mandatory education on generative AI, alone or as an alternative to a small fine (Gauthier, In re R.A., Rosslyn2016, Booker). Suspension from practice before the court (Lnu v Blanche). Public reprimand (Johnson v Dunn). And referral to the licensing body, which appears in 48 rows and is examined in its own section below.
What it cost
Forty-five rows state a figure, in seven currencies. The table below sums each currency separately; the report does not convert, because exchange rates would add a false precision to sums that already mix penalties, fees and reimbursements.
| Currency | Rows with a figure | Sum | Median | Largest | Largest row |
|---|---|---|---|---|---|
| USD | 34 | 371,542 | 5,000 | 110,204 | Couvrette v Wisnovsky (D. Or., 12 Dec 2025) |
| CAD | 4 | 36,000 | 7,500 | 20,000 | Arbuckle v Tanner (N.S.C.A., 12 Aug 2026) |
| ILS | 2 | 36,000 | 18,000 | 30,000 | Ben Cohen v Ramat Gan (Israel S.C., 22 Mar 2026) |
| GBP | 1 | 24,727.20 | 24,727.20 | 24,727.20 | Bandla v SRA (EWHC Admin, 13 May 2025) |
| AUD | 2 | 17,469.23 | 8,734.62 | 9,097.93 | Asif v Minister for Immigration (FedCFamC2G, 22 Jul 2026) |
| SGD | 1 | 10,000 | 10,000 | 10,000 | Tan Hai Peng v Tan Cheong Joo (SGHC, 6 Mar 2026) |
| EUR | 1 | 1,250 | 1,250 | 1,250 | ’s-Hertogenbosch 2026:93 (27 Jul 2026) |
The US series is the only one long enough to show a trend, and the trend is in the tail, not the middle.
| Year | US rows with a figure | Sum (USD) | Median (USD) | Largest (USD) |
|---|---|---|---|---|
| 2023 | 1 | 5,000 | 5,000 | 5,000 (Mata v Avianca) |
| 2024 | 3 | 14,000 | 2,000 | 10,000 (Kruse v Karlen) |
| 2025 | 11 | 174,454 | 5,000 | 110,204 (Couvrette v Wisnovsky) |
| 2026 to 3 Sep | 19 | 178,088 | 5,000 | 46,511 (Kleyman Law Group v Kaloidis) |
One booking convention affects the year table: Couvrette’s USD 94,704 fee award was fixed on 23 March 2026 but is recorded against the 12 December 2025 sanctions order, so USD 94,704 of the 2025 total was awarded in 2026. On the order date alone the 2025 total is USD 79,750 and the 2026 total USD 272,792.
The USD 5,000 that Judge Castel ordered in Mata v Avianca on 22 June 2023 is still the median order three years later. Sixteen of the 34 US figures are below USD 5,000, seven sit between USD 5,000 and USD 9,999, ten between USD 10,000 and USD 49,999, and one above USD 50,000. What has changed is how often the tail appears: no order above USD 10,000 exists before Lacey v State Farm in May 2025, and six have followed in the sixteen months since.
| Band (USD) | Rows | Rows |
|---|---|---|
| Under 1,000 | 1 | In re R.A. (Iowa, USD 150 or two hours of CLE) |
| 1,000 to 4,999 | 15 | Smith v Farwell, Gauthier, Dehghani, Benjamin v Costco, Kadince (Utah), Shahid v Esaam, State v Coleman, Del Biaggio, Ledoux, Mendel S., Tov Realty, Barteca, In re BFI, Capital Standard, Adams |
| 5,000 to 9,999 | 7 | Mata, Wadsworth, Coomer, Mid Central, Lnu v Blanche, NM Supreme Court (Aarons, unverified), Booker v Kroger |
| 10,000 to 49,999 | 10 | Kruse, Noland, Deutsche Bank v LeTennier, Cassata, Ibach, Full Standard Properties, Scott, Rosslyn2016, Lacey, Kleyman |
| 50,000 and above | 1 | Couvrette v Wisnovsky |
Two distinctions the headline sums hide. First, penalty versus fees. Couvrette’s USD 110,204 is a USD 15,500 sanction plus USD 94,704 in fees awarded to the defendants across two orders; the press figure of “USD 110,000” is the aggregate. Kleyman’s USD 46,511 is a USD 10,000 sanction plus USD 36,511.35 in fees and costs. Rosslyn2016’s USD 29,877 is entirely fees payable to the Chapter 7 trustee, ordered jointly against counsel and the petitioning creditors. Full Standard Properties’ USD 14,500 is a reimbursement counsel paid before the show-cause hearing, after which Judge Freeman discharged the order. Lacey’s USD 31,100 is a costs award against two firms. Of the eleven orders at USD 10,000 or above, eight are wholly or mainly fee-shifting (Lacey, Couvrette, Ibach, Kruse, Rosslyn2016, Full Standard Properties, Kleyman and Cassata). The largest penalty payable to a court in the dataset is the USD 15,500 ordered payable to the Clerk of the District of Oregon in Couvrette; the largest per-item penalty is USD 15,000, set at USD 1,500 for each of ten false items in Scott v Illinois Human Rights Commission, where the Appellate Court of Illinois said that courts have no choice but to increase fines for AI-hallucinated citations until they have a significant deterrent effect.
Second, per-lawyer versus per-order. Several rows aggregate penalties across people. Wadsworth’s USD 5,000 is USD 3,000 against the drafting lawyer and USD 1,000 against each of two signing lawyers. Coomer’s USD 6,000 is USD 3,000 against each of two lawyers. Deutsche Bank v LeTennier’s USD 10,000 is USD 7,500 against counsel and USD 2,500 against the client. Cassata’s USD 10,000 is USD 1,000 each against an associate and her supervising partner, payable to the Lawyers’ Fund for Client Protection, plus USD 8,000 in fees against the firm. Lnu v Blanche’s USD 5,000 is USD 2,500 against each of two lawyers, alongside the six-month suspension. A per-lawyer median would be lower than USD 5,000.
Seven rows carry an education order: Gauthier v Goodyear (a CLE course on generative AI, with the USD 2,000 penalty), Dastou v Holmes, In re R.A. (two hours of AI-ethics CLE in lieu of the USD 150), Rosslyn2016 (six hours of generative-AI CLE), Tov Realty v Suarez, LPC Commercial Services v Ar-Razzaaq Properties and Booker v Kroger (proof of ethics and technology CLE). The education order is the one remedy courts have invented for this conduct rather than borrowed from the general sanctions toolkit.
The escalation curve
The ledger reads as four stages. Each is dated to the decision that marks it, and each decision is a tracker row with its primary document.
2023: the template. On 22 June 2023 Judge P. Kevin Castel of the Southern District of New York sanctioned two lawyers USD 5,000 in Mata v Avianca for citing six ChatGPT-invented decisions and persisting after they were questioned. Five other 2023 rows follow the same shape without money: a Texas appeal in Ex parte Lee, two federal litigants in person, a South African regional court in Parker v Forsyth (unverified) and the first UK tribunal decision, Harber v HMRC, where the First-tier Tribunal dismissed the appeal and imposed nothing.
2024: the first referral, the first regulator. Of 11 rows, four carry money (USD 2,000 in Smith v Farwell, USD 10,000 against a self-represented appellant in Kruse v Karlen, USD 2,000 plus a CLE course in Gauthier v Goodyear, the first order to name Claude). The Second Circuit’s referral of counsel to its Grievance Panel in Park v Kim (30 January 2024) is the first regulator referral in the ledger, and Zhang v Chen (20 February 2024) the first Canadian one, with a Law Society of British Columbia investigation. The year’s most consequential row carries no penalty at all: Dayal (27 August 2024), in which an Australian family-court judge referred a solicitor to the Victorian regulator over four non-existent authorities generated by the AI feature of his practice-management software.
2025: the tail appears. Forty rows. Lacey v State Farm (6 May 2025) is the first order above USD 10,000 against lawyers: USD 31,100 against Ellis George and K&L Gates after a privilege brief built on undisclosed AI research was filed, queried, and re-filed with the fabrications intact. Johnson v Dunn (23 July 2025) is the first public reprimand and disqualification of large-firm lawyers, with no fine, the court finding fines an insufficient deterrent. Noland v Land of the Free (12 September 2025) is the first California appellate sanction, USD 10,000, with the opinion served on the State Bar. Ko v Li is the first referral for criminal contempt: after the lawyer admitted, in a letter of 30 September 2025, that she had lied about who drafted a ChatGPT-fabricated factum, the court referred a contempt proceeding to the Attorney General (the tracker dates the second ruling, 2025 ONSC 6785, to 4 December 2025; the Law Society Tribunal’s reasons record the referral as ongoing). And Couvrette v Wisnovsky (12 December 2025) is the largest matter in the ledger: fifteen non-existent cases across three summary-judgment briefs, amended briefs that deleted the quotation marks rather than the fabrications, terminating sanctions, USD 15,500 in sanctions, USD 94,704 in fees and a referral to the Oregon State Bar. Outside the United States, the Divisional Court in Ayinde v Haringey (6 June 2025) set the England and Wales position, referring a barrister to the Bar Standards Board and solicitors to the Solicitors Regulation Authority while stating that the contempt threshold could have been met.
2026: courts reach for their own disciplinary powers. Ninety-three rows to 3 September. The first US suspension came from an appellate court rather than a bar: the Ninth Circuit in Lnu v Blanche (3 June 2026) suspended two lawyers from practice before it for six months and fined each USD 2,500 after they described AI fabrications as typographical errors. The Supreme Court of Alabama in Ibach v Stewart (24 April 2026) dismissed the appeal, ordered USD 17,200 plus double appellate costs and barred counsel from further unsigned filings. The Appellate Court of Illinois in Scott (28 July 2026) set a per-item rate. The Supreme Court of New York, Kings County, in Kleyman (4 August 2026) found 23 fictitious decisions and 83 misstated real ones across 19 filings in a lawyer’s own fee suit. In the same eight months the first bar-level suspensions were reported in Nebraska, Ontario and Minnesota; only the Ontario decision has been read in the tribunal’s own reasons.
The curve is therefore two curves. The median order has been flat at USD 5,000 for three years. The maximum, the number of orders above USD 10,000, the use of suspension and the frequency of referral have all risen, and 2026 added a new instrument, the appellate court’s own disciplinary jurisdiction, that does not wait for a bar.
Counsel, litigants in person and public bodies
The actor field is coded on 79 of 150 rows: 52 lawyers, 21 litigants in person, 4 adjudicators or officials coded “other”, 1 judge and 1 firm. The 71 uncoded rows are mostly 2023 to mid-2026 entries added before the field existed, so proportions below are indicative only.
| Principal outcome | Lawyer | Litigant in person | Adjudicator, official, judge or firm | Not coded |
|---|---|---|---|---|
| Sanctions | 19 | 3 | 0 | 15 |
| Warning | 9 | 10 | 0 | 8 |
| Fine | 8 | 0 | 0 | 2 |
| Other | 5 | 3 | 5 | 14 |
| Referral | 5 | 0 | 1 | 13 |
| Costs order | 2 | 3 | 0 | 6 |
| Suspension | 2 | 0 | 0 | 1 |
| Dismissal | 1 | 1 | 0 | 11 |
| Pending | 1 | 1 | 0 | 1 |
Among coded rows, lawyers draw the money and litigants in person draw the warnings: 10 of 21 litigants in person received a warning as the principal outcome, and only 3 received a monetary order, none of them in US dollars (AP v SK, CAD 1,000; Asif, AUD 9,097.93; Arbuckle v Tanner, CAD 20,000, where the appellant filed AI-generated transcripts with invented testimony and a forged transcriber’s certificate). The uncoded set contains at least one large order against a self-represented party, the USD 10,000 in Kruse v Karlen, so the pattern is a tendency rather than a rule. Where US courts have ordered litigants in person to pay, it has usually been the opponent’s fees rather than a penalty, as in Varma v Bank of New York Mellon, where the California Court of Appeal ordered respondent’s appellate fees in an amount to be set by the trial court and sent the opinion to the State Bar over possible unauthorised practice.
Among lawyers with a US dollar figure (19 rows), the median is USD 3,000 and the sum USD 175,488; the larger uncoded set of 15 rows sums to USD 196,054 because it contains Couvrette, Lacey and Kruse. Recoding the older rows is the first item in this report’s change list.
Five rows record a decision-maker, not an advocate, as the source of the fabricated authority. Two are Indian tax adjudicators whose orders the Supreme Court of India set aside in Pooja Ramesh Singh (2 July 2026) and Gadiya v Union of India (2 September 2026, an INR 4.25 billion customs penalty); one is an acting judge in Johannesburg referred to the Legal Practice Council in FJL v TGO (31 July 2026); and Ben Cohen v Ramat Gan (22 March 2026) is the one row in the tracker in which a court ordered money against a public body, NIS 30,000 against a municipality that relied on a non-existent ministry circular, as recorded in the tracker from the Supreme Court of Israel’s decision and confirmed in two English-language reports of it. In none of the five did a court impose a personal penalty on the official; the Indian court left discipline to the appointing authority, and the South African referral is pending.
Which tools the orders name
| AI product named in the record | Rows |
|---|---|
| Not named | 92 |
| ChatGPT (alone or with others) | 27 |
| GPT-4o, named as such | 1 |
| Generative AI, product unspecified or suspected (one row lists ChatGPT, Gemini, Copilot and CoCounsel only as the court’s examples) | 14 |
| Claude (alone or with others) | 5 |
| Google product (Bard, Gemini or AI search) | 4 |
| Westlaw AI tools or CoCounsel (one row in which counsel denied ChatGPT and named Westlaw’s tools) | 3 |
| Microsoft Copilot | 2 |
| Practice-management or specialist tool (LEAP, Visto.ai, Legal Genius, Strongsuit, ChatOn, a firm’s internal tool, Open Law) | 7 |
The lines sum to 63 for 58 rows because four orders name more than one product and are counted on each line: Lacey names CoCounsel, Westlaw Precision and Google Gemini from a sworn declaration; Ledoux v Outliers names ChatGPT and Claude; Barteca v Tacobarn names Open Law, then Claude and ChatGPT; JNE24 names Claude for research and Copilot for validation. In 92 of 150 rows (61 percent) the order does not identify the product, most often because the court did not need to: the finding is that the authority does not exist, whatever generated it. Singapore’s High Court made the point explicit in Tan Hai Peng, holding that citing fictitious authorities is sanctionable whether or not AI produced them and ordering SGD 5,000 each against the counsel on record and the supervising solicitor.
The tool count should not be read as a comparison of products. It measures which products lawyers admitted using in a filing that went wrong, weighted heavily toward the one most people use, and it says nothing about the error rates of any of them. That question has its own report, How often does legal AI hallucinate?.
After the courtroom: what regulators did
Forty-eight rows carry a regulatoryOutcome. The table sorts them by what the record actually shows.
| What the record shows | Rows | Examples |
|---|---|---|
| Referral or notice to a regulator, nothing further on record | 38 | Park v Kim (2d Cir. Grievance Panel, Jan 2024); Ayinde (BSB and SRA, Jun 2025); Couvrette (Oregon State Bar, Dec 2025); Ibach (Alabama State Bar, Apr 2026); Lnu v Blanche (State Bar of California, Jun 2026); Scott (Illinois ARDC, Jul 2026); Kleyman (NY Grievance Committee, Aug 2026); three Brazilian bar notices; two Eleventh Circuit committee referrals |
| Disposition verified from the regulator’s own document | 5 | Dayal (VLSB+C practising certificate varied, 19 Aug 2025); OBA v Reeves (Texas 4 Nov 2025, Tennessee 15 Dec 2025, Alabama 6 Mar 2026, Oklahoma 27 May 2026, all recited in the Oklahoma opinion); Law Society of Ontario v Lee, 2026 ONLSTH 136 (six-month suspension from 17 Jul 2026 and CAD 10,000 costs, decision 16 Jul 2026, read from a mirrored copy of the tribunal’s reasons; the Ko v Li row records the same disposition); In re Mitchell (USPTO public reprimand, final order 27 Jul 2026); ‘s-Hertogenbosch (formal reprimand and EUR 1,250 costs, 27 Jul 2026) |
| Disposition reported but not read in a primary document, or investigation without recorded outcome | 4 | In re Faisal Ahmed (30-day suspension, Minnesota, order not located); Prososki v Regan (reported suspension of counsel, Nebraska, not confirmed from any primary document); Noland (California State Bar Court recommendation reported); Zhang v Chen (Law Society of British Columbia investigation recorded, outcome not on record) |
| Regulator action predated the conduct | 1 | Bandla v SRA (the appellant had already been struck off; the fabricated authorities appeared in his appeal against that decision) |
The five verified dispositions are worth reading for what regulators did when they did act. The Law Society Tribunal of Ontario accepted a joint submission and suspended the lawyer in Ko v Li for six months from 17 July 2026 with CAD 10,000 in costs, for misconduct that included misleading the court and the Law Society about who had drafted the factum. The Victorian Legal Services Board and Commissioner did not suspend the Dayal solicitor; on 19 August 2025 it varied his practising certificate so that he may practise only as an employee, may not handle trust money or run his own practice, and must complete two years of supervised practice with quarterly reporting. The bars of Texas, Tennessee, Alabama and Oklahoma each publicly reprimanded or censured Matthew Reeves for the five ChatGPT citations in Johnson v Dunn, a reciprocal chain that ran from 4 November 2025 to 27 May 2026 and is the only multi-state disciplinary sequence in the ledger. The USPTO’s Office of Enrollment and Discipline publicly reprimanded a patent attorney by settlement for an AI-drafted claim-construction chart whose citations to the patent’s own record were inaccurate. The Dutch council rejected a mere warning, found breaches of the core values of expertise and integrity, and ordered EUR 1,250 in costs.
Time from referral to disposition, where it can be measured:
| Matter | Court decision | Regulator decision | Days |
|---|---|---|---|
| Johnson v Dunn to Texas reprimand of Reeves | 23 Jul 2025 | 4 Nov 2025 | 104 |
| Johnson v Dunn to Oklahoma reciprocal reprimand | 23 Jul 2025 | 27 May 2026 | 308 |
| Dayal referral to VLSB+C variation | 27 Aug 2024 | 19 Aug 2025 | 357 |
| Ko v Li first contempt ruling (2025 ONSC 2965) to Law Society Tribunal decision | 20 May 2025 | 16 Jul 2026 | 422 |
Every other referral in the ledger is younger than the slowest of these, or has produced nothing that the tracker can find. The 38 open referrals include every referral to the Solicitors Regulation Authority (Ayinde, UK v SSHD, Cork v Smith), the Bar Standards Board (Ayinde, MS Bangladesh, Lodhia), the State Bar of California (Noland’s recommendation aside), The Florida Bar, the North Carolina State Bar, the Maryland Attorney Grievance Commission, the Oregon State Bar, the Alabama State Bar (Ibach), the New York grievance committees, the Legal Practice Council of South Africa (three matters) and the Brazilian bar (three matters). This is the follow-up list. The tracker’s watch-list carries each with its referral date.
Two self-reports belong in this section because they are the regulator layer working without a referral. In Cork v Smith (22 May 2026) Pinsent Masons quoted a fabricated Insolvency Rule to the High Court after a junior solicitor used an AI system, self-referred to the SRA and agreed to pay the clients’ costs of the correspondence; ICC Judge Mullen’s judgment is the public record and the SRA investigation is open. In Sullivan & Cromwell (18 April 2026) the firm told the bankruptcy court what had happened; the row carries no penalty and no referral.
Outside the United States
Sixty-one rows sit outside US courts, in thirteen tracker jurisdictions. The United Kingdom has 17, Canada 10, India 8, Australia 6, South Africa 5, Brazil 4, Israel 3, Ireland 3, and one each for New Zealand, the Netherlands, Qatar’s financial-centre court, Singapore and China.
The UK series is the largest non-US series and the least monetary. Of 17 rows, one states a figure: GBP 24,727.20 in indemnity costs in Bandla v SRA, against a struck-off solicitor appearing in person who denied using AI. Eleven of the 17 are dismissals, warnings or “other”, most of them tribunal decisions involving litigants in person. The professional-conduct weight sits in the referrals: Ayinde (BSB and SRA), MS (Bangladesh) (Upper Tribunal to the BSB), UK v SSHD (a solicitor and compliance officer to the SRA over supervision of a caseworker who used Google AI), Lodhia v Twelve Trees (direct-access counsel to the BSB) and Cork v Smith. None has a published SRA or BSB disposition as of 5 September 2026.
Canada’s 10 rows include the ledger’s two largest non-US costs orders against individuals, CAD 20,000 in Arbuckle v Tanner and the CAD 10,000 costs with the Lee suspension, and the only criminal-contempt referral (Ko v Li). Australia’s 6 rows include the ledger’s first regulator disposition (Dayal) and two immigration matters with costs orders (JNE24, AUD 8,371.30, referred to the Legal Practice Board of Western Australia; Asif, AUD 9,097.93 against an applicant in person). India’s 8 rows are almost all decisions set aside rather than penalties. South Africa’s 5 include three Legal Practice Council referrals with no disposition, one of them the acting judge in FJL v TGO. Singapore’s one row, Tan Hai Peng, is the clearest statement in any jurisdiction that the supervising lawyer is liable alongside the drafter.
Timeline
| Date | Decision | What it added to the ledger |
|---|---|---|
| 22 Jun 2023 | Mata v Avianca (S.D.N.Y.) | USD 5,000; the template and the still-current median |
| 4 Dec 2023 | Harber v HMRC (FTT) | First UK decision; no penalty |
| 30 Jan 2024 | Park v Kim (2d Cir.) | First regulator referral (Grievance Panel) |
| 20 Feb 2024 | Zhang v Chen (BCSC) | First Canadian decision; costs and a law-society investigation |
| 27 Aug 2024 | Dayal (FedCFamC2F) | First referral that produced a verified regulator disposition (19 Aug 2025) |
| 25 Nov 2024 | Gauthier v Goodyear (E.D. Tex.) | First order naming Claude; first CLE order |
| 6 May 2025 | Lacey v State Farm (C.D. Cal.) | First order above USD 10,000 against lawyers (USD 31,100) |
| 6 Jun 2025 | Ayinde v Haringey (Div. Ct.) | England and Wales position; BSB and SRA referrals |
| 23 Jul 2025 | Johnson v Dunn (N.D. Ala.) | First public reprimand and disqualification; start of the four-state reciprocal chain |
| 4 Nov 2025 | Texas reprimand of Reeves | First verified US bar disposition |
| 4 Dec 2025 (per the tracker) | Ko v Li (ONSC) | First criminal-contempt referral, recorded as ongoing in the Law Society Tribunal’s reasons |
| 12 Dec 2025 | Couvrette v Wisnovsky (D. Or.) | Largest matter (USD 110,204 aggregate); terminating sanctions |
| 6 Mar 2026 | Tan Hai Peng (SGHC) | Supervising solicitor held liable alongside counsel |
| 22 Mar 2026 | Ben Cohen v Ramat Gan (Israel S.C.) | First costs order against a public body (NIS 30,000) |
| 24 Apr 2026 | Ibach v Stewart (Ala.) | Appeal dismissed, USD 17,200, double costs, filing bar |
| 27 May 2026 | OBA v Reeves (Okla.) | Fourth reciprocal reprimand closes the chain |
| 3 Jun 2026 | Lnu v Blanche (9th Cir.) | First suspension, by a court’s own disciplinary power |
| 2 Jul 2026 | Pooja Ramesh Singh (SCI) | Tribunal order built on fake authority is no decision in law |
| 16 Jul 2026 | Law Society of Ontario v Lee | First law-society suspension (six months, from 17 Jul 2026) |
| 27 Jul 2026 | In re Mitchell (USPTO); ‘s-Hertogenbosch 2026:93 | First patent-office and first EU bar dispositions |
| 28 Jul 2026 | Scott v Illinois HRC (Ill. App.) | Per-item rate of USD 1,500 |
| 4 Aug 2026 | Kleyman v Kaloidis (N.Y. Sup.) | Largest 2026 order (USD 46,511); 23 fictitious decisions across 19 filings |
| 2 Sep 2026 | Gadiya v Union of India (SCI) | INR 4.25 billion penalty order set aside |
What to watch
The 38 open referrals are the report’s forward calendar. The oldest with no disposition is Park v Kim (30 January 2024). The ones most likely to produce a public decision next, on the pattern of the four that have, are the state-bar referrals of 2025 (Couvrette to Oregon; Noland’s California recommendation, if adopted; Mezu to Maryland) and the SRA and BSB matters from Ayinde, now fifteen months old. Two reported suspensions (Nebraska, Minnesota) need their primary orders; when those are retrieved the verified-disposition count moves from five to seven.
Two open court processes could change the top of the ledger. The criminal-contempt referral in Ko v Li remains unresolved before the Ontario Superior Court. The reconsideration history in Couvrette closed in June 2026 with the sanctions intact, but the Oregon State Bar file is open.
Three rules due or recent will change how future rows are coded. Ireland’s Practice Direction HC 142 (in operation 1 September 2026) requires a declaration in affidavits and witness statements, so Irish rows may start to carry certification breaches. The SRA’s warning notice on misuse of AI (17 August 2026) gives its investigators a document to cite in the open SRA matters. And the EU AI Act’s Article 50 transparency duties, applicable from 2 August 2026 under the Commission’s guidelines, were not needed by the Dutch council but may be by a later one.
Three sentences journalists can quote
As of 5 September 2026 the SafeLegalAI incident tracker records 150 court and regulator decisions on AI-fabricated authority in fifteen jurisdictions, 45 of them with a stated monetary order; the 34 United States orders total USD 371,542 with a median of USD 5,000.
Forty-eight of the 150 decisions referred the conduct to a professional regulator, and five of those referrals have produced a disposition that can be read in the regulator’s own document: in Victoria, Australia; in a four-state reciprocal chain in the United States; before the Law Society Tribunal of Ontario; at the US Patent and Trademark Office; and before a Dutch bar disciplinary council.
The largest matter in the record is Couvrette v Wisnovsky (D. Or.), USD 15,500 in sanctions ordered 12 December 2025 plus USD 94,704 in fees fixed 23 March 2026; the largest per-item penalty is USD 15,000 in Scott v Illinois Human Rights Commission (Ill. App. 1st, 28 July 2026), set at USD 1,500 per false item.
Appendix A: every row with a stated monetary order
Figures as recorded in the order. “Type” is the report’s reading of the order: penalty (payable to a court or fund), fees (payable to an opponent or trustee), costs (a costs order), or mixed. Unverified rows are marked.
| Date | Row | Court | Amount | Type |
|---|---|---|---|---|
| 22 Jun 2023 | Mata v Avianca | S.D.N.Y. | USD 5,000 | penalty |
| 12 Feb 2024 | Smith v Farwell | Mass. Super. Ct. | USD 2,000 | penalty |
| 13 Feb 2024 | Kruse v Karlen | Mo. Ct. App. | USD 10,000 | damages for frivolous appeal |
| 25 Nov 2024 | Gauthier v Goodyear | E.D. Tex. | USD 2,000 | penalty, plus CLE |
| 24 Feb 2025 | Wadsworth v Walmart | D. Wyo. | USD 5,000 | penalty (3,000 + 1,000 + 1,000) |
| 2 Apr 2025 | Dehghani v Castro | D.N.M. | USD 1,500 | penalty |
| 24 Apr 2025 | Benjamin v Costco | E.D.N.Y. | USD 1,000 | fine |
| 6 May 2025 | Lacey v State Farm | C.D. Cal. (Special Master) | USD 31,100 | costs |
| 13 May 2025 | Bandla v SRA | EWHC (Admin) | GBP 24,727.20 | indemnity costs |
| 22 May 2025 | Kadince (Utah Ct. App.; tracker row filed under the appellant’s name) | Utah Ct. App. | USD 1,000 | mixed |
| 28 May 2025 | Mid Central Operating Engineers v HoosierVac | S.D. Ind. | USD 6,000 | penalty |
| 30 Jun 2025 | Shahid v Esaam | Ga. Ct. App. | USD 2,500 | penalty |
| 7 Jul 2025 | Coomer v Lindell | D. Colo. | USD 6,000 | penalty (3,000 × 2) |
| 15 Aug 2025 | JNE24 v Minister for Immigration | FedCFamC2G | AUD 8,371.30 | costs |
| 12 Sep 2025 | Noland v Land of the Free | Cal. Ct. App. | USD 10,000 | penalty |
| 1 Oct 2025 | In re R.A. | Iowa Ct. App. | USD 150 | penalty, or two hours of CLE |
| 1 Oct 2025 | Specter Aviation v Laprade (unverified) | Quebec | CAD 5,000 | fine |
| 12 Dec 2025 | Couvrette v Wisnovsky | D. Or. | USD 110,204 | mixed: 15,500 penalty ordered 12 Dec 2025 (ECF 215); 94,704 fees fixed 23 Mar 2026 (ECF 225), booked to the sanctions-order date |
| 8 Jan 2026 | Deutsche Bank v LeTennier | N.Y. App. Div. 3d | USD 10,000 | penalty (7,500 counsel + 2,500 client) |
| 27 Jan 2026 | Cassata v Macrina | N.Y. Sup. Ct. | USD 10,000 | mixed (1,000 × 2 to Lawyers’ Fund + 8,000 fees) |
| 6 Mar 2026 | Tan Hai Peng v Tan Cheong Joo | SGHC | SGD 10,000 | personal costs (5,000 × 2) |
| 20 Mar 2026 | State v Coleman | Ohio Ct. App. | USD 2,000 | fine |
| 22 Mar 2026 | Ben Cohen v Ramat Gan | Israel S.C. | ILS 30,000 | costs against a municipality |
| 24 Apr 2026 | Ibach v Stewart | Ala. | USD 17,200 | fees and costs to the appellee, plus double appellate costs payable to the court |
| 3 Jun 2026 | Lnu v Blanche | 9th Cir. | USD 5,000 | penalty (2,500 × 2), plus six-month suspension |
| 10 Jul 2026 | Del Biaggio v Bansen | Cal. state court | USD 1,500 | fine |
| 14 Jul 2026 | In re Rosslyn2016 | Bankr. S.D. Tex. | USD 29,877 | fees to the trustee, plus six hours of CLE |
| 16 Jul 2026 | Law Society of Ontario v Lee | Law Society Tribunal | CAD 10,000 | costs, with six-month suspension |
| 22 Jul 2026 | Asif v Minister for Immigration | FedCFamC2G | AUD 9,097.93 | costs |
| 24 Jul 2026 | Ledoux v Outliers | US federal court | USD 3,000 | penalty |
| 27 Jul 2026 | ’s-Hertogenbosch 2026:93 | Raad van Discipline | EUR 1,250 | disciplinary costs |
| 28 Jul 2026 | Scott v Illinois HRC | Ill. App. 1st | USD 15,000 | penalty (1,500 × 10) |
| 31 Jul 2026 | Mendel S. v Faigy S. | N.Y. state court | USD 1,000 | penalty |
| 31 Jul 2026 | Tov Realty v Suarez | Conn. | USD 2,000 | penalty (1,000 each from counsel and the firm, to the CT Bar Institute) |
| 4 Aug 2026 | Barteca v Tacobarn | D. Conn. | USD 3,500 | penalty |
| 4 Aug 2026 | Kleyman Law Group v Kaloidis | N.Y. Sup. Ct., Kings | USD 46,511 | mixed (10,000 penalty + 36,511.35 fees and costs) |
| 6 Aug 2026 | In re BFI Waste Systems | US federal court | USD 1,500 | penalty |
| 11 Aug 2026 | AP v SK | Que. C.A. | CAD 1,000 | penalty |
| 12 Aug 2026 | Arbuckle v Tanner | N.S.C.A. | CAD 20,000 | lump-sum costs |
| 17 Aug 2026 | Haifa Labor Court 7669-06-24 | Haifa Labor Court | ILS 6,000 | costs |
| 20 Aug 2026 | Full Standard Properties v SJ 6070 | N.D. Cal. | USD 14,500 | fees reimbursed before the hearing |
| 21 Aug 2026 | Capital Standard v US Bank | Fla. state court | USD 1,500 | fine |
| 21 Aug 2026 | NM Supreme Court (Aarons) (unverified) | N.M. | USD 5,000 | penalty (press report) |
| 27 Aug 2026 | Adams v Matrix Providers | US federal court | USD 1,000 | fine |
| 28 Aug 2026 | Booker v Kroger | N.D. Ga. | USD 8,000 | penalty to the sanctions fund, plus CLE |
Appendix B: sources
Primary documents for the rows this report relies on most, each linked from its tracker page and re-read for this report on 5 September 2026:
- Mata v Avianca, Inc., No. 1:22-cv-01461 (S.D.N.Y.), Opinion and Order on Sanctions, 22 June 2023: CourtListener docket
- Kruse v Karlen, No. ED111172 (Mo. Ct. App. E.D. 13 Feb 2024): opinion PDF
- Gauthier v Goodyear Tire & Rubber Co., No. 1:23-CV-281 (E.D. Tex. 25 Nov 2024): CourtListener docket
- Dayal [2024] FedCFamC2F 1166: AustLII
- Wadsworth v Walmart Inc., No. 2:23-cv-00118 (D. Wyo. 24 Feb 2025): CourtListener docket
- Lacey v State Farm General Insurance Co., No. 2:24-cv-05205 (C.D. Cal.), Order of Special Master, ECF 119: RECAP PDF
- Bandla v Solicitors Regulation Authority [2025] EWHC 1167 (Admin): Find Case Law
- Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank, Divisional Court, 6 June 2025: judiciary.uk PDF
- Johnson v Dunn, No. 2:21-cv-01701-AMM (N.D. Ala. 23 Jul 2025), sanctions order ECF 204: RECAP PDF
- Noland v Land of the Free, L.P., No. B331918 (Cal. Ct. App. 2d Dist. 12 Sep 2025): California Courts PDF
- In the Interest of R.A., No. 24-1629 (Iowa Ct. App. 1 Oct 2025): CourtListener
- Ko v Li, 2025 ONSC 2965 (20 May 2025) and 2025 ONSC 6785 (4 Dec 2025): CanLII 2965, CanLII 6785 (CanLII blocks scripts; the 20 May 2025 decision was also read from a mirrored PDF on 5 September 2026)
- Couvrette v Wisnovsky, No. 1:21-cv-00157-CL (D. Or.), sanctions opinion ECF 215 and fees opinion ECF 225: RECAP 215, RECAP 225
- Deutsche Bank Natl. Trust Co. v LeTennier, 2026 NY Slip Op 00040 (App. Div. 3d Dept. 8 Jan 2026): CourtListener
- Cassata v Michael Macrina Architect, P.C., 2026 NY Slip Op 26014 (Sup. Ct. Suffolk Cty. 27 Jan 2026): CourtListener
- Tan Hai Peng Micheal v Tan Cheong Joo [2026] SGHC 49 (6 Mar 2026): eLitigation; Singapore Law Watch PDF
- State v Coleman, 2026-Ohio-965 (Ohio Ct. App. 11th Dist. 20 Mar 2026): CourtListener
- Prososki v Regan, 321 Neb. 38, No. S-25-295 (Neb. 20 Mar 2026): opinion PDF, Nebraska advance sheets via CourtListener storage; Justia case page
- Ben Cohen v Municipality of Ramat Gan, AAM 63194-08-25 (Israel S.C. 22 Mar 2026): Supreme Court of Israel (the host does not answer script requests; the NIS 30,000 and the 22 March date are as recorded in the tracker row and reported by law.co.il and Pearl Cohen, both by the same authors; law.co.il gives 23 March)
- Ibach v Stewart, No. SC-2025-0106 (Ala. 24 Apr 2026): CourtListener
- Cork & Anor v Smith [2026] EWHC 1199 (Ch): Find Case Law
- State ex rel. Oklahoma Bar Association v Reeves, 2026 OK 37 (27 May 2026), paragraphs 1, 5, 6, 7 and 24 for the Texas (4 Nov 2025), Tennessee (15 Dec 2025) and Alabama (6 Mar 2026) dispositions: CourtListener; Justia (read in a browser on 5 September 2026; both hosts block scripts)
- Lnu v Blanche, No. 24-4790 (9th Cir. 3 Jun 2026), disciplinary order: CourtListener
- In re Rosslyn2016, LLC, No. 25-34507, ECF 312 (Bankr. S.D. Tex. 14 Jul 2026): Charlotin archive PDF
- Law Society of Ontario v Lee, 2026 ONLSTH 136 (16 Jul 2026): CanLII (blocks scripts); tribunal reasons read from a mirrored PDF on 5 September 2026
- In re Brian E. Mitchell, USPTO Proceeding No. D2026-16, final order 27 Jul 2026: USPTO OED PDF
- Raad van Discipline ‘s-Hertogenbosch, ECLI:NL:TADRSHE:2026:93 (27 Jul 2026): tuchtrecht.overheid.nl
- Scott v Illinois Human Rights Commission, 2026 IL App (1st) 251462 (28 Jul 2026): Illinois Courts PDF
- FJL v TGO, Case No. 2025/220239 (Gauteng Division, Johannesburg, full court, 31 Jul 2026): Charlotin archive PDF
- Kleyman Law Group, P.C. v Kaloidis, Index No. 502644/2025 (Sup. Ct. Kings Cty. 4 Aug 2026): Charlotin archive PDF
- Arbuckle v Tanner, 2026 NSCA 62 (12 Aug 2026): Charlotin archive PDF
- Full Standard Properties v SJ 6070, No. 5:25-cv-09451, ECF 41 (N.D. Cal. 20 Aug 2026): RECAP PDF
- In re Faisal S. Ahmed (Minnesota), MPR News, 21 Aug 2026 (press report; the Supreme Court order has not been located): mprnews.org
- Varma v The Bank of New York Mellon, No. E085566 (Cal. Ct. App. 27 Aug 2026): Charlotin archive PDF
- Booker v The Kroger Co., No. 1:26-cv-02006-SDG, ECF 66 (N.D. Ga. 28 Aug 2026): RECAP PDF
- Vijay Ghanshyam Gadiya v Union of India, 2026 INSC 947 (2 Sep 2026): Supreme Court of India PDF
- Global baseline: Damien Charlotin, AI Hallucination Cases database, read 4 September 2026 (2,016 entries): damiencharlotin.com/hallucinations
- SafeLegalAI data: Incident Tracker · statistics · JSON export · methodology · related reports: Three years of legal AI hallucinations, Legal AI incident report, Q3 2026
Appendix C: changes to this report
None since first publication. Planned for the next edition: recode actor on the 71 rows that predate the field; retrieve the primary orders for the Nebraska, Ontario and Minnesota suspensions; add a per-lawyer penalty series alongside the per-order series.