The best-documented AI hallucination examples come from courts, because judges write down what happened. Lawyers in New York and Alabama filed ChatGPT-invented case citations and were sanctioned, and a Canadian tribunal held an airline responsible for its chatbot's wrong answer. Published studies show the same pattern in news answers and speech transcription. Each case below links to its primary source.
01What makes an example documented?
Every case here rests on a primary record: a court order, a tribunal decision or a published study. We use case names as the court recorded them and leave out personal details. News stories alone were not enough for inclusion.
This page is general information, not legal advice. If a court filing or contract matters to you, ask a lawyer.
02Mata v. Avianca: invented court opinions (New York, 2023)
In a personal injury case in the US District Court for the Southern District of New York, the plaintiff's lawyers filed a brief that cited court opinions which did not exist. The judge's 22 June 2023 sanctions order says the fake opinions, with fake quotes and citations, came from ChatGPT.
The order is careful on one point: using an AI tool is not improper in itself. The court said the rules make lawyers the gatekeepers of what they file. It also found that the lawyers kept defending the opinions after the court questioned them, and it found bad faith by the individual lawyers. The penalty was $5,000.
The order lists the harms: the other side spends money exposing the fakes, the court loses time that other cases needed, the client may lose arguments built on real precedent, and judges falsely named as authors of fake opinions can suffer damage to their reputation.
The record includes screenshots in which the lawyer asked ChatGPT if one case was real. It answered that the case was real and could be found in legal databases.
03Johnson v. Dunn: a second sanctions order (Alabama, 2025)
Two years later, a federal judge in the Northern District of Alabama dealt with five false citations in two motions filed in a prison civil rights case. In her 23 July 2025 opinion, the court said the citations were hallucinated by ChatGPT.
The court found that one lawyer acted with recklessness tantamount to bad faith by signing motions containing citations another lawyer had added. It decided that a fine and a public reprimand alone were not enough. It publicly reprimanded three lawyers, disqualified them from the case and referred the matter to the Alabama State Bar.
Compare the two orders. The 2023 penalty was a $5,000 payment. The 2025 order removed the lawyers from the case. The two courts chose different responses.
04Moffatt v. Air Canada: a chatbot gives the wrong policy (Canada, 2024)
A customer asked an airline's website chatbot about bereavement fares and was told he could apply for the discount after travelling. Another page on the same site said otherwise. The British Columbia Civil Resolution Tribunal, in decision 2024 BCCRT 149, found the airline liable for negligent misrepresentation, according to a law firm summary.
The airline argued that the chatbot was a separate entity responsible for its own actions. The tribunal rejected that and treated the chatbot as part of the airline's website, noting that it makes no difference whether information comes from a static page or a chatbot.
This is a small-claims tribunal decision, so it does not bind other courts. It is still a clear example of a company answering for what its own chatbot said.
05Examples from research: news answers and speech transcription
In the EBU and BBC study, journalists reviewed more than 3,000 answers about the news from four AI assistants. Of all answers, 45% had at least one significant issue, and 20% had major accuracy problems, including invented details and outdated information.
In the Careless Whisper study, researchers tested OpenAI's Whisper speech-to-text system as it stood in 2023. About 1% of transcriptions contained phrases that were never spoken, and roughly 38% of those invented phrases caused concrete harm, such as violent language or false links. Invented content was more frequent for speakers with longer pauses in speech.
Counts of court cases keep growing. When we read the public AI hallucination cases database on 24 October 2026, it listed 2,207 decisions and called itself a work in progress.
06What the cases teach
The same few lessons appear in every case:
- The person or company that publishes the output is answerable for it, whether the text came from a chatbot or not.
- Asking the chatbot if its answer is real does not work. In New York it confirmed a case that did not exist.
- Tools that retrieve documents first make fewer errors but still make them, as a Stanford test of legal research products found (Magesh and others).
- Check names, citations and quotes in the original source before you file, send or sign anything.
?What is not known
- The sources do not say how many invented citations are caught before filing, so the cases above are only those that became public.
- Courts are still deciding how severe sanctions for fabricated citations should be, and rules about disclosing AI use in filings vary by court.
- The Whisper rate comes from a 2023 version of the system. Newer versions may behave differently.
QQuestions people ask
Were the lawyers punished for using AI?
The New York court said using a reliable AI tool is not improper in itself. The sanctions were for filing false citations and, in that case, for standing by them after being questioned.
Can a company be held responsible for its chatbot's mistakes?
In the Air Canada tribunal decision, yes, for that customer's claim. It is a small-claims ruling and does not bind other courts.
Last checked October 24, 2026. Rules and settings change, so open the source to see the current version.
- Mata v. Avianca, Opinion and Order on Sanctions US District Court, Southern District of New York The court's 22 June 2023 order; we used its findings, the chatbot screenshots it describes and the $5,000 penalty.
- Johnson v. Dunn, 792 F.Supp.3d 1241 US District Court, Northern District of Alabama (reported copy hosted by Berkeley Law) The 23 July 2025 opinion; we used its description of the five citations and the sanctions imposed.
- Moffatt v. Air Canada (2024 BCCRT 149) commentary McCarthy Tétrault LLP A law firm summary of the tribunal decision; we used it for the facts, the holding and the rejected argument.
- News Integrity in AI Assistants press release European Broadcasting Union The 22 October 2025 release on a study led by the BBC and EBU; we used its headline percentages.
- Careless Whisper: Speech-to-Text Hallucination Harms arXiv (Koenecke and others) A 2024 research paper on invented phrases in speech transcription; we used its rate, its harm share and its finding on speech pauses.
- AI Hallucination Cases database Damien Charlotin A public tracker of court decisions involving fabricated AI content; we used its case count as read on 24 October 2026.
- Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools arXiv (Magesh and others) A preregistered Stanford test of legal research tools; we used its finding that retrieval does not remove errors.
General information, not legal, medical, financial or investment advice. We are not affiliated with any company, agency or standards body named here. See our editorial policy and corrections, or tell us about a mistake at [email protected].