AI Hallucination Sanctions Cases: What Courts Have Actually Done

Courts across the United States have sanctioned lawyers for filing briefs citing cases that do not exist. Penalties have escalated from four-figure fines in 2023 to $15,000 per attorney, six-month suspensions, and disciplinary referrals. What draws the heaviest sanction is rarely the AI use itself — it is the failure to check, and the lack of candor afterward.

This post is the case law. For the rule, read our guide to Rule 11 sanctions; for the mechanism, read why language models fabricate citations.

This article reflects decisions and reporting available as of September 2026. This area moves fast, and several matters below are pending.

How many of these cases are there?

More than anyone tracks precisely, because not every incident produces a written order.

The reference point practitioners use is Damien Charlotin's AI Hallucination Cases database, a public tracker of decisions in which a court identified hallucinated material in a filing. Its inclusion rule is conservative — only what a court found or clearly implied, never what a party merely alleged. "I am not making that judgment," Charlotin has explained; "I let the courts and judges make or imply it."

Charlotin noted in February 2026 that new entries were arriving at roughly five per day, and in a June 2026 survey Norton Rose Fulbright counted more than 1,148 documented cases, noting the Fifth Circuit's observation that the problem "shows no sign of abating."

Mata v. Avianca: the case everything is measured against

Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) (S.D.N.Y. June 22, 2023), is the foundational decision. Two attorneys at Levidow, Levidow & Oberman submitted an opposition brief containing six nonexistent judicial opinions with fabricated quotations, produced by ChatGPT — then defended them after the court questioned whether the cases existed.

Judge P. Kevin Castel drew the line every later court has followed:

Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.

The order imposed a $5,000 penalty jointly and severally on Peter LoDuca, Steven A. Schwartz, and the firm, under Rule 11 or, alternatively, inherent authority, plus letters to their client and to each judge falsely identified as the author of a fabricated opinion. What the court found sanctionable was not the ChatGPT query but "conscious avoidance and false and misleading statements" — the conduct after the filing.

What courts have done in 2026

The Ninth Circuit: suspension enters the range

Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026) is the year's most consequential order, and the panel published it. Attorneys Mike Singh Sethi and William Rounds filed immigration briefs containing nonexistent cases, misattributed quotations, and gross misrepresentations of real authority. They called the errors typographical and repeatedly denied that generative AI might be responsible.

The panel was explicit that AI use was not the offense:

The rules are not violated at the point of research and drafting, but at the point of signing and filing.

It was equally explicit about what earned the discipline: "The gravity of discipline we impose, including the temporary suspension of practice, is owed to this repeated failure of candor." Had the attorneys disclosed the AI use in their motion to correct and apologized for failing to check the brief, the court said, "lesser sanctions may have been warranted."

The order imposed $2,500 on each attorney personally; a six-month suspension from Ninth Circuit practice; service of the order on their clients, opposing counsel, and the presiding judge in every pending case; a two-year requirement that firm filings carry a sworn statement naming any AI tool used and certifying personal review of every citation; and service on the State Bar of California.

Nebraska: the appeal itself was the casualty

In Prososki v. Regan, No. S-25-0295, a dissolution-of-marriage appeal argued to the Nebraska Supreme Court in February 2026, the appellant's brief contained fabricated case names alongside fabricated quotations attributed to real Nebraska decisions. The court struck the brief, dismissed the appeal, and referred counsel to the state's Counsel for Discipline.

Counsel denied using generative AI. The court's reasoning did not turn on the answer: filing fictitious authority breaches the rules of professional conduct whether a machine or a person produced it. The disciplinary proceeding is pending; local reporting in April 2026 described an interim suspension meanwhile. The client's appeal was decided without a brief — the cost practitioners most underweight.

Western District of Washington: it spreads to the experts

The Tacoma case LeDoux v. Outliers, Inc. produced a sequence of orders. In July 2026 the court sanctioned plaintiff's counsel for "dozens of inaccurate factual and legal citations across at least five different filings." In August, per the Washington State Bar Association, it excluded a plaintiff's expert whose report contained hallucinated citations, holding the opinions unreliable under Rule 702 because the fabrications "shatter … his credibility with this Court." Without the expert, summary judgment went to the defense. The exposure is not confined to lawyers: anything you file that someone else drafted carries your certification.

California: a state supreme court steps in

In January 2026 the California Supreme Court granted review in Kjoller v. Superior Court, S293723, and directed the Third District Court of Appeal to issue an order to show cause why sanctions should not be imposed on the Nevada County District Attorney over briefing alleged to contain fabricated authority; per Horvitz & Levy, the court may appoint a referee to take evidence. The district attorney acknowledged "errored citations" but maintained they "belong to real cases." The matter is pending and no sanction has issued. See also California's AI ethics rules for lawyers.

The pattern: what aggravates, what mitigates

Courts are not punishing tool selection. They are punishing what happened between the fabrication and the court finding it.

Aggravating, in rough order of severity:

1. Denial — attributing a nonexistent case to a typo. The Ninth Circuit said candid disclosure at the motion-to-correct stage might have drawn a lesser sanction.
2. Repetition across filings. One bad brief is an error. Five is a practice, and the Washington court said so.
3. Filing again after a warning. The ABA Journal reported a Southern District of Ohio matter in which Senior Judge Walter H. Rice found the most egregious Rule 11 violations he had seen on the bench: $7,500 against two attorneys, a contempt finding, and referral to Ohio disciplinary counsel.
4. Refusing to engage with the show-cause order. In Whiting, counsel declined to answer the Sixth Circuit's seven questions and challenged the order itself — $15,000 each, plus fees and double costs.
5. Corrections that introduce new errors.

Mitigating: prompt unprompted admission, self-reporting before opposing counsel finds it, concrete remediation, and naming the tool. Gamez v. County of Fresno ended with the show-cause order discharged after counsel gave a candid, detailed account. Concealment converts a Rule 11 problem into a candor problem, and candor problems produce suspensions.

Mitigation is not immunity. In United States v. Farris, counsel responded promptly and candidly and was still removed from the case, denied all Criminal Justice Act compensation, and referred to two chief judges and the Kentucky Bar Association. Sterne Kessler's 2025 review reached the same conclusion: courts sanctioned attorneys who denied AI use more harshly than those who disclosed it.

The rules courts are invoking

No court has needed a new rule. Rule 11(b) is the workhorse: your signature certifies the legal contentions are warranted by existing law after a reasonable inquiry, a citation to a nonexistent case fails that on its face, and under Rule 11(c)(1) the firm is ordinarily jointly responsible. 28 U.S.C. § 1927 reaches the vexatious multiplication that defending a fabricated citation causes, and inherent authority covers what Rule 11's procedure does not. On appeal, FRAP 46 authorizes suspension or disbarment from the court's bar and discipline for conduct unbecoming — what made the Ninth Circuit suspension possible.

The professional-conduct rules run alongside: competence (Model Rule 1.1, whose comment 8 reaches the risks of relevant technology), candor toward the tribunal (Model Rule 3.3), and supervisory duties. ABA Formal Opinion 512 applies all of them to generative AI, and most states have followed — see our state-by-state bar AI guidance.

The duty, stated plainly

Signing a filing means you read what it cites and confirmed the authority says what you say it says. Practically, that means opening the case — not the summary, not the snippet, not the paraphrase. Our walkthroughs on verifying AI-generated citations and cite-checking a brief cover the mechanics, including the failure mode that trips people up most: the real case, correctly cited, that does not stand for the proposition attached to it. Every court above sanctioned that variety alongside the invented ones.

This is where the research tool matters. A general-purpose chatbot generates text that looks like citations. A retrieval-based system like CaseRead searches your firm's case files and the public law together and answers from sources it actually returned, flagging what it cannot verify. That does not remove your verification duty — nothing does — but it changes what you are verifying from an assertion into a document. Whether disclosure is separately required is jurisdiction-specific: see do lawyers have to disclose AI use.

If it happens to you

1. Stop filing. Pull anything unfiled that shares the same research.
2. Verify the whole document, not the flagged citation. Fabrications travel in groups.
3. Correct in writing, immediately, and state where the error came from — including that AI produced it. Do not call a nonexistent case a typo.
4. Notify opposing counsel and the court before either raises it.
5. Answer the show-cause order completely. Refusing to engage produced the largest per-attorney sanctions above.
6. Check your reporting obligations, and fix the process. Courts credit concrete remediation over apology.

Frequently asked questions

How many AI hallucination sanctions cases are there?

No one has an authoritative count — not every incident produces a written order. The most-cited public tracker is Damien Charlotin's AI Hallucination Cases database, which logs only decisions where a court found or clearly implied that a filing relied on hallucinated material. Charlotin noted in February 2026 that new entries were arriving at roughly five per day.

Can a lawyer be suspended for filing AI-fabricated citations?

Yes. In Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026), the Ninth Circuit fined two attorneys $2,500 each and suspended both from practice before that court for six months, ordered two years of sworn AI-disclosure certifications from their firm, and served the order on the State Bar of California. Suspensions remain the exception.

Does using AI to draft a brief violate Rule 11?

No. Courts have been consistent that the tool is not the violation. The Ninth Circuit put it directly: the rules are not violated at the point of research and drafting, but at the point of signing and filing. What Rule 11 sanctions is filing a paper you did not verify.

What should I do if I filed a brief with a fabricated citation?

Correct it immediately, in writing, and say plainly where the error came from — including that AI produced it. Notify opposing counsel and the court before either finds it. Do not characterize a nonexistent case as a typographical error. Courts have discharged show-cause orders where counsel gave a candid account, and escalated sharply where counsel did not.

---

Every case above began with a filing nobody opened before it was signed. If a draft in front of you came out of a general-purpose AI tool, run it through the Hallucination Shield first. It is free, no signup, and each citation it finds — up to 25 per run — is checked for existence and support. That is cheaper than a show-cause order.

This article is for general information and is not legal advice. Several matters above are pending; confirm current status before relying on any decision cited here.