Attorneys Sanctioned for AI & Unverified Citations: What the Courts Are Doing
Published June 2026 · Updated June 2026
Courts across the United States are now sanctioning attorneys who file briefs containing fabricated or unverified case citations — many of them generated by AI tools such as ChatGPT. The penalties range from monetary fines to disqualification and disciplinary referral. The rule the courts keep returning to is simple: the lawyer who signs a filing is accountable for the citations in it.
The problem, by the numbers (2026)
- 1,350+
- court decisions worldwide that identified AI-fabricated or unverified citations — a lower-bound count from Damien Charlotin’s public tracker (as of June 2026).
- 17 in one day
- separate U.S. court decisions flagged suspected AI hallucinations on a single day — March 31, 2026 — as the pace of orders accelerated.
- $5K → $10K+
- monetary sanctions have climbed from $5,000 in Mata v. Avianca (2023) to $10,000 in California’s first published AI-citation opinion (Noland, 2025), and higher since.
Catalogue counts from the public AI-hallucination case tracker maintained by Damien Charlotin.
The trend, in one sentence
What began as a single embarrassing incident in 2023 has become a steady stream of published sanction orders. Generative AI will confidently produce citations that look real but do not exist — “hallucinations” — and judges have made clear that “the AI wrote it” is not a defense. Rule 11 of the Federal Rules of Civil Procedure (and its bankruptcy counterpart, Rule 9011) places the duty of reasonable inquiry on the signing attorney.
Notable cases and recent orders
17 hallucination decisions in a single day
March 31, 2026Legal commentators documented 17 U.S. court decisions issued on one day, each noting suspected AI-generated hallucinations in court filings — a snapshot of how routine these orders have become in 2026.
Source: Reason / Volokh Conspiracy.
Noland v. Land of the Free, L.P. — California’s first published AI-citation opinion ($10,000)
Cal. Ct. App., Sept. 2025In California’s first published opinion on AI-hallucinated citations, the Court of Appeal (2d Dist., Div. 3, B331918) sanctioned attorney Amir Mostafavi $10,000 and referred him to the State Bar after finding that 21 of 23 case quotations in his opening brief were fabricated by generative AI and never appeared in any published case.
Source: McGuireWoods.
Boston v. Williams, No. 1:23-cv-00752 (N.D. Ga.) — Rule 11 sanctions
N.D. Ga., Oct. 2025On October 28, 2025, Judge Ray sanctioned plaintiffs’ counsel under Rule 11 after defendants showed that most of the cases cited in a summary-judgment opposition — by one count, 17 of 24 — were inaccurate or did not exist. Counsel conceded the brief had been AI-generated and filed without review.
Source: Frankfurt Kurnit (technologylaw).
Bankruptcy court — AI citations held to violate Rule 9011
November 2025A U.S. bankruptcy judge found that filings containing AI-fabricated citations violated Federal Rule of Bankruptcy Procedure 9011 — the bankruptcy analog of Rule 11 — confirming the duty of candor follows attorneys into every forum.
Source: fedcivilprocedure.com.
Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023) — the case that started it
S.D.N.Y., June 2023Judge P. Kevin Castel sanctioned two attorneys $5,000 after they submitted a brief citing six cases that ChatGPT had invented — and then stood by the fake authorities when challenged. It remains the reference point for every AI-citation sanction that followed.
Source: CourtListener docket.
Case summaries above are drawn from the linked reporting and court records. Always confirm the underlying order against the court’s official docket before relying on it — the same discipline this page is about.
The legal duty: Rule 11 and Rule 9011
By signing a pleading, motion, or other paper, an attorney certifies — after a reasonable inquiry — that its legal contentions are warranted by existing law. Courts have held that presenting fabricated authority without that inquiry can violate Rule 11 (and Rule 9011 in bankruptcy); sanctions follow notice and an opportunity to respond. Separate duties of candor and competence under state professional-conduct rules, and a court’s inherent authority, can apply independently. Penalties have included monetary fines, orders to notify clients and opposing counsel, mandatory CLE, and referral to disciplinary authorities.
How to avoid sanctions
- Confirm the case exists. Look the citation up in an authoritative source — the official reporter, the court’s docket, CourtListener, or PACER. If you cannot find it, it likely does not exist.
- Read the actual opinion. Open the opinion and read the cited passage yourself. Do not rely on an AI summary of what the case says.
- Check the quotation and holding. Confirm the quotation is exact and that the holding actually supports the proposition you are citing it for.
- Verify the citation details. Check the reporter, court, and year against the primary record so the citation format is correct.
- Keep AI as a lead, not a source. Treat AI output as a research starting point only. Nothing reaches a filing until a human has read the underlying source.
Where RulingIQ fits
RulingIQ is a free judicial-research platform built on public U.S. court records, under a strict no-fabrication standard: we never invent statistics. When a figure cannot be measured from real data, we show “insufficient data” instead of a placeholder number. That is the same discipline that keeps fabricated citations out of a filing — rely on verified primary sources, and refuse to present anything you cannot stand behind.
Frequently asked questions
- Have attorneys actually been punished for using AI citations?
- Yes — and increasingly so. Beginning with Mata v. Avianca (S.D.N.Y. 2023), where two lawyers were fined $5,000 for citing ChatGPT-invented cases, courts have imposed monetary sanctions, disqualifications, and disciplinary referrals on attorneys who filed briefs with AI-hallucinated or unverified citations. A public tracker maintained by researcher Damien Charlotin had identified more than 1,350 such court decisions worldwide by June 2026, spanning federal trial and appellate courts, state appellate courts, and bankruptcy courts.
- What rule do these sanctions fall under?
- Most arise under Federal Rule of Civil Procedure 11 (and its bankruptcy analog, Rule 9011), which requires the attorney who signs a filing to certify — after a reasonable inquiry — that its legal contentions are warranted. Separate duties of candor and competence under state professional-conduct rules, and a court’s inherent authority, can apply on their own. The recurring theme: if your signature is on the document, you are accountable for the citations in it.
- Why does generative AI produce fake case citations?
- Large language models generate text by predicting plausible word sequences, not by retrieving verified records. When asked for supporting authority, a model can produce citations that look correct — real-sounding case names, reporters, and pin cites — but do not exist or do not say what the model claims. These are called "hallucinations," and they cannot be trusted without independent verification against a primary source.
- How can I avoid being sanctioned?
- Verify every citation against a primary source (the official reporter, the court’s docket, or an authoritative database such as CourtListener or PACER) before filing. Confirm the case exists, that the quotation is accurate, and that it actually supports your proposition. Never file authority you have not personally read. Treat AI output as a research lead, never as a citation of record.
- How does RulingIQ help?
- RulingIQ is a judicial-research platform built on public U.S. court records and an explicit no-fabrication standard: we never invent statistics, and any figure we cannot measure from real data is shown as "insufficient data" rather than a made-up number. Researching your judge and the underlying records on verified data is part of the same discipline that keeps fabricated citations out of your filings.