The best public count is Damien Charlotin's AI Hallucination Cases database. As of its 2 September 2026 update it lists 2,008 court decisions worldwide, 1,378 of them in the USA. Only 213 US entries carry any monetary sanction. The largest US penalty on record is $110,204 in Couvrette v. Wisnovsky, D. Oregon, 23 March 2026.
The database, and what it actually counts
There is one credible public tracker of AI errors reaching courts: the AI Hallucination Cases database maintained by Damien Charlotin at damiencharlotin.com/hallucinations. It has been cited in news media and, according to its own front page, in several court decisions dealing with hallucinated material.
As of its last update on 2 September 2026, the database lists 2,008 cases identified worldwide (AI Hallucination Cases database). The USA accounts for 1,378 of them. Canada is second with 214, Australia third with 110, and the UK fourth with 62.
Read the scope note before quoting the number. The database tracks "legal decisions in cases where generative AI produced hallucinated content," and it says plainly: "It does not track the (necessarily wider) universe of all fake citations or use of AI in court filings" (AI Hallucination Cases database).
This page is not legal advice. It is a read of a public dataset, downloaded and counted on 3 September 2026.
Where the internet is commonly wrong: cases are not people
Almost every article that cites this database converts its count into a headcount of sanctioned professionals. That conversion is wrong in three separate ways.
First, the unit is a court decision, not a person. One decision can involve several lawyers at one firm. One lawyer can appear across several decisions. The 2,008 figure is neither an upper nor a lower bound on individuals.
Second, most entries carry no money at all. Of the 1,378 US entries, only 213 record any monetary sanction, and 185 of those record a known amount rather than a placeholder for an unknown one. That is roughly 15 percent of US entries with a monetary penalty attached.
Third, and most consequential: most US entries are not professionals. Pro se litigants account for 817 of the 1,378 US entries. Lawyers account for 533. Judges account for 4, experts 10, prosecutors 2, paralegals 1, and one federal defender. When a headline says "hundreds of professionals sanctioned," it has silently absorbed several hundred self-represented litigants.
The US penalty leaderboard, current as of 3 September 2026
| Case | Court | Date | Party | Penalty |
|---|---|---|---|---|
| Couvrette v. Wisnovsky | D. Oregon | 23 March 2026 | Lawyer | $110,204 |
| ByoPlanet International v. Johansson and Gilstrap | S.D. Florida | 1 August 2025 | Lawyer | $85,567 |
| LiveVideo.AI Corp. v. Redstone | S.D. New York | 9 June 2026 | Lawyer | $80,056 |
| Minjie Zheng v. ICANN | C.D. California | 23 December 2025 | Pro se litigant | $66,129 |
| Howell Management Services, LLC v. Vashisht-Rota | CA California | 15 December 2025 | Pro se litigant | $64,235 |
| Shana Jordan et al. v. Chicago Housing Authority et al. | CC Illinois | 5 December 2025 | Lawyer | $59,500 |
| Heimkes v. Fairhope Motorcoach Resort COA | S.D. Alabama | 31 March 2026 | Lawyer | $55,597 |
| Kleyman Law Group, P.C. v. Kaloidis | SC New York | 4 August 2026 | Lawyer | $46,511 |
| Joel A. Rivera v. Triad Properties Corporation et al. | N.D. Alabama | 31 March 2026 | Lawyer | $35,603 |
| Lacey v. State Farm General Insurance | C.D. California | 6 May 2025 | Lawyer | $31,100 |
All figures above are as recorded in the database's downloadable CSV, retrieved 3 September 2026 (AI Hallucination Cases database).
The largest US penalty, and what it was actually made of
Couvrette v. Wisnovsky in the District of Oregon, dated 23 March 2026, is the largest US figure in the database at $110,204. The composition matters. The database records the outcome as "Briefs struck; Monetary sanction (15.5k); Adverse costs order (94.7k); claims dismissed with prejudice."
So the sanction proper was about $15,500. The other $94,700 was an adverse costs order — the other side's legal fees. That pattern repeats across the top of the table. Courts are recovering the opponent's wasted expense far more often than they are levying large punitive fines.
The practical lesson is worth stating plainly. The cost of an AI citation error is usually not the fine. It is the fees of everyone who had to prove the citation was fake, plus the case outcome. In Couvrette the claims were dismissed with prejudice. That is the more expensive half.
Two numbers the internet still quotes that are now stale
"$31,100 is the largest AI sanction." Lacey v. State Farm General Insurance, C.D. California, 6 May 2025, imposed $31,100 jointly and severally against two law firms. It was widely reported as a record. It now sits tenth on the US list.
"Mata v. Avianca was a huge fine." The 2023 case that started the genre, Mata v. Avianca, Inc. in the Southern District of New York, dated 22 June 2023, carried a $5,000 monetary fine against the lawyers and firm, per the database. It was consequential for the publicity, not the amount.
Discipline is rarer than sanctions, and rarer than people assume
Money is one consequence. A referral to a bar disciplinary body is another, and it is the one that threatens a license.
Of the 1,378 US entries, 112 record a professional sanction or disciplinary referral, and 1,266 record none (AI Hallucination Cases database, CSV retrieved 3 September 2026). That is about 8 percent. Several of the largest-penalty cases carry both: ByoPlanet was referred to the Florida Bar, and LiveVideo.AI, Heimkes and Kleyman all record bar referrals alongside their monetary awards.
The database records court outcomes. It is not a register of what those bars later decided. Whether a referral became a suspension, a reprimand or nothing at all is not tracked here, and we could not verify a comprehensive public source that does track it.
The trend line, and the naming problem
The growth is steep. US entries by year run 11 in 2023, 37 in 2024, 526 in 2025 and 804 in 2026 through early September (AI Hallucination Cases database, CSV retrieved 3 September 2026). Worldwide the same years run 16, 59, 850 and 1,083.
Which tool caused it is mostly unknown. Across US entries, 970 record the AI tool as "Implied" and 251 as "Unidentified." Only 60 name ChatGPT, 4 name Claude, 3 name Perplexity, 3 name CoCounsel and 3 name Copilot. The database itself warns that naming a tool "does not necessarily mean that tool was responsible for the hallucinations in question."
Anyone telling you which product is worst is reading noise. In more than 88 percent of US entries, the court never established which tool was used.
Beyond law, the honest answer is that nobody is counting
This database covers courts and tribunals. It does not cover accountants, doctors, engineers, financial advisers or the far larger population of employees disciplined internally for AI errors that never reached a filing.
We could not verify any comparable public database for those professions. Anyone quoting a total for "professionals sanctioned for AI mistakes" across all fields is extrapolating from a legal dataset that explicitly does not support the claim. The honest position is that the legal profession is the only one where the numbers are public, and that is because court filings are public, not because lawyers are worse.
How to check these numbers yourself
- Open damiencharlotin.com/hallucinations. Read the "Last updated" date near the top. Everything below it moves weekly.
- Read the scope note first. It defines what counts as a case and what does not.
- Filter by State to USA. The country filter shows the count next to each name.
- Filter by Party. Compare "Lawyer" against "Pro Se Litigant" before you quote any total as professional misconduct.
- Click "Download CSV." The table view truncates. The CSV carries the full penalty, outcome and professional sanction fields.
- Treat a penalty value of 1 as a placeholder. The site states that 1 means a monetary sanction was imposed but the amount is unknown, and that these are excluded from its own penalty graphs and averages.
- Read the underlying order before citing a case. Most rows link to the source document.
sources:
- Damien Charlotin — AI Hallucination Cases database (2,008 cases; last updated 2 September 2026; CSV retrieved 3 September 2026): https://www.damiencharlotin.com/hallucinations/
- AI Hallucination Cases — CSV export: https://www.damiencharlotin.com/hallucinations/hallucinations/download.csv
- Lacey v. State Farm General Insurance, C.D. Cal., 6 May 2025 (order via the database): https://www.damiencharlotin.com/documents/311/Lacey_v._State_Farm_General_Insurances_Co._D._Cal._May_6_2025.pdf
- ABA Formal Opinion 512 (context on lawyers' verification duties): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
related:
- Can lawyers use ChatGPT and keep client confidentiality?
- Is it safe to put my client data into ChatGPT?
- How do I verify an AI-generated citation before I file it?
- What happens if my accountant or adviser uses AI and gets it wrong?
