AI Hallucination Sanctions in 2026: 13 Verified Cases in One Table

Last updated: 3 November 2026. This table is built from court opinions and orders, ABA Journal, Bloomberg Law, Above the Law, the Law Society Gazette and The Global Legal Post, with total counts taken from Damien Charlotin’s AI Hallucination Cases database. It is a selection, not a complete list. We name cases and courts, not individual lawyers, and we flag where a court did not find that AI was used.

Short answer: Nobody can list every 2026 sanction for AI hallucinations: one database alone holds 2,262 decisions worldwide. This table tracks 13 notable rulings we verified against court orders or major legal press. The largest total is about $110,000, in Couvrette v. Wisnovsky, while the harshest outcomes are suspensions, dismissals and a default judgment.

Key takeaways

• Charlotin’s database listed 2,262 decisions worldwide, 1,533 of them from the United States, when it was last updated on 2 November 2026. Many are minor, and many involve people without lawyers.
• The biggest totals come from fee-shifting, not fines. Couvrette reached about $110,000 because of opposing counsel’s fees; the fine was about $15,500.
• Money is rarely the worst of it. Lawyers in this table lost cases for clients, lost admission to courts for two years, or were suspended.
• Courts punish the cover-up as much as the error. Denials, evasive answers and poor supervision appear again and again as aggravating factors.

Why can nobody list every AI sanction?

The honest answer is scale and definition. Most sanctions sit in unpublished orders or in trial court dockets, and new ones arrive daily. The best public record is the AI Hallucination Cases database, kept by researcher Damien Charlotin, which grew from 2,006 cases on 1 September 2026, the figure in our earlier piece on what the sanctions record tells lawyers, to 2,262 two months later. It records decisions where a court or tribunal has addressed suspected AI-fabricated material, whether or not it imposed a penalty.

So this article does something narrower. It picks 13 rulings from 2026 that stand out for the amount, the court or the type of penalty, and we checked each against the opinion or a major legal publisher.

Notable verified AI hallucination sanctions in 2026

Two notes on the table. Couvrette’s fine dates from December 2025, and we include it because the fee order that took the total above $100,000 was made in 2026. And in Whiting, the lawyers did not answer the court’s question about AI, so the panel made no finding on it, although it discussed how AI hallucinations arise.

What is the largest AI hallucination sanction in 2026?

Among the cases we verified, it is Couvrette v. Wisnovsky, a family dispute over an Oregon winery. ABA Journal reported about $80,000 in fees and about $15,000 in fines against the lead lawyer, and $14,000 in fees against local counsel. Magistrate Judge Mark Clarke called the case “a notorious outlier in both degree and volume”. The filings included 15 references to nonexistent cases and eight fabricated quotations. Fortune described it as the largest such penalty in American legal history, but with thousands of decisions on file, treat that as a report, not a certainty. Nobody admitted using AI in the ABA Journal’s account: the defendants said the client probably drafted the briefs with it, and the local lawyer said he did not write or review them.

For pure fines, the appellate courts now lead. In Whiting, the Sixth Circuit explained why it went to $15,000 each: “smaller fines have plainly been inadequate”. The Illinois Appellate Court took a formula approach in Scott, and the Oregon Court of Appeals fined a lawyer $10,000 over 15 fabricated citations.

What do courts punish most?

The cases show four patterns, and none of them is really about technology.

• Lack of candour. In Fletcher, the Fifth Circuit found that a lawyer’s answers to its show-cause order were misleading about the source of the errors. In Lnu, the Ninth Circuit sanctioned two lawyers partly for a lack of candour about the errors. In the Nebraska matter, the lawyer told the justices he had not used AI, then admitted it in a later affidavit, according to local reports.
• Failed supervision. In Gentry, a suit against New Orleans officials, the supervisor paid four times what the junior lawyer did. In Cork v Smith, the judge criticised two supervising solicitors over a junior’s heavy reliance on AI.
• Signing what you did not read. Local counsel in Withers were fined and disqualified. In Couvrette, local counsel paid 15% of the fees even though he said he had no hand in the briefs.
• Repetition. Flycatcher ended in default judgment after repeated filings with fake citations. Judge Katherine Polk Failla called the sanctions “extraordinary”, ABA Journal reported.

The Fifth Circuit summed up the mood, calling hallucinations a problem that is getting worse rather than better. Fletcher also says what a lawyer using generative AI must do: “ensure that the legal propositions and authority generated are trustworthy.”

Who actually pays when a filing contains fake citations?

Often, the client. Three rows in the table ended the client’s case: default judgment in Flycatcher, dismissal with prejudice in Couvrette and a dismissed appeal in Ibach v. Stewart, where the Alabama Supreme Court called the misuse “widespread and particularly egregious”. In Withers, reported by Bloomberg Law, both sides lost their lawyers and the trial was cancelled.

The firm can pay too. In Cork v Smith, Pinsent Masons covered its former client’s costs and apologised, according to the Law Society Gazette. The case shows that a large firm is not immune.

What power do courts use to sanction AI errors?

Underneath the sanction sits professional ethics. In the US, ABA Formal Opinion 512 treats AI as a matter of competence, confidentiality, candour and supervision. In England and Wales, the High Court’s Ayinde decision in 2025 set the tone, and Cork v Smith shows that a junior’s error can become a firm-level problem.

How can a law firm avoid being in this table?

Telling lawyers to check citations has not stopped the problem. A safer approach builds the check into the process:

• Pull every cited authority from a primary law database before filing, and read it. Do not ask the same tool that drafted the text to confirm it.
• Make the signing lawyer personally confirm the check, including local counsel on a pro hac vice filing.
• Have a written rule on which tools may be used and for what. Our law firm AI policy guide gives a structure.
• If an error surfaces, correct it fast and say what happened. The cases above show that the second answer often does more harm than the first mistake.
• Choose tools built on licensed content for research. Our comparison of AI legal research platforms explains the difference, and the Legal AI software directory lists the options.

How did we choose the cases?

Each row had to meet three tests: a decision or key order dated in 2026, a named court and case, and an outcome stated in the order itself or by a major legal publisher. We left out rulings where we could confirm the amount only from AI vendor blogs or aggregators, including several late 2026 orders in the Charlotin database. We did not include warnings with no penalty, and we did not list cases against people representing themselves, though they make up a large share of the database. We will add rows as more orders are reported.

Frequently asked questions

How many AI hallucination cases are there?

Damien Charlotin’s database listed 2,262 decisions worldwide, including 1,533 from the United States, when it was updated on 2 November 2026. It counts court and tribunal decisions that deal with suspected AI fabrications, and not all of them ended in a sanction. The true number of incidents is unknown, because many errors are never caught.

What is the biggest fine for AI hallucinations?

The largest total we verified is about $110,000 in Couvrette v. Wisnovsky in Oregon, and most of that was fee-shifting. The largest direct fines in our table are $15,000, in Whiting v. City of Athens, per lawyer, and in Scott v. Illinois Human Rights Commission. Larger orders may exist that we have not verified.

Can a lawyer be suspended for using AI?

Not for using AI as such. Lawyers are disciplined for filing false authority and for failing to be candid. In 2026 the Ninth Circuit suspended two lawyers from practising before it for six months, and the Nebraska Supreme Court temporarily suspended a lawyer whose brief it was examining.

Can clients lose their case because of a lawyer’s AI errors?

Yes. In Flycatcher, the client suffered a default judgment. In Couvrette, the case was dismissed with prejudice, and in Ibach v. Stewart the Alabama Supreme Court dismissed the appeal. A client in that position may have a claim against the lawyer, but that is a separate fight.

Does this happen in the UK?

Yes. In Cork v Smith [2026] EWHC 1199 (Ch), the judge publicly admonished Pinsent Masons and two solicitors over two misleading letters. The firm referred itself to the Solicitors Regulation Authority, and the judge decided that contempt proceedings would be disproportionate. English judges have warned lawyers about this since Ayinde in 2025.

Do lawyers have to tell the court they used AI?

It depends on the court. There is no single national rule in the US. More than 300 federal and state judges require some form of AI disclosure, EDRM reported in April 2026, so check the local rules and the judge’s own orders for every filing. Whatever the rule on disclosure, the lawyer who signs remains responsible for every citation.

This article is general information on reported court decisions as of 3 November 2026. It is not legal advice. Some orders may be under appeal or changed after reporting. Allegations about AI use are described only as far as the court or the cited publication reported them.