ChatGPT-invented court cases in Mata v. Avianca

ChatGPT-invented court cases in Mata v. Avianca
	FALSE
	"Court decisions including Varghese v. China Southern Airlines, Shaboon v. Egyptair and Martinez v. Delta Air Lines are real legal precedents supporting a passenger's lawsuit against an airline."
Domain	Technology & Media
Tactic	Fabrication;
Era	Current
First seen	2023
Origin	A legal brief filed in March 2023 in the US District Court for the Southern District of New York, drafted with the help of ChatGPT
Last reviewed	2026-10-26
	Further reading: books on Technology & Media

The ChatGPT-invented court cases in Mata v. Avianca became the best-known early example of artificial intelligence "hallucination" causing real-world harm. In 2023, lawyers representing a man suing the airline Avianca filed a brief in a New York federal court that cited at least six court decisions that did not exist. The cases, complete with names, citations, quotations and internal references, had been generated by ChatGPT. When the airline's lawyers and the judge could not find them, the lawyers asked ChatGPT whether the cases were real, and it assured them they were. The judge sanctioned the lawyers, and the case became a warning to professionals in every field about relying on generative AI without checking its output.

The claim

The plaintiff, Roberto Mata, said he had been injured when a metal serving cart struck his knee on an Avianca flight from El Salvador to New York in 2019. Avianca asked the court to dismiss the case because it had been filed too late under the time limits of the Montreal Convention, the international treaty governing airline liability. In response, Mata's lawyers filed a brief arguing that the case should proceed and citing a series of decisions that appeared to support their position.

Among the cases cited were Varghese v. China Southern Airlines, Shaboon v. Egyptair, Petersen v. Iran Air, Martinez v. Delta Air Lines, Estate of Durden v. KLM Royal Dutch Airlines and Miller v. United Airlines. The brief included quotations from these decisions and citations to the volumes in which they were supposedly reported.

Origin

The brief was prepared by Steven A. Schwartz, a lawyer with about thirty years of experience at the New York firm Levidow, Levidow & Oberman. Because Schwartz was not admitted to practice in that federal court, his colleague Peter LoDuca signed and filed the papers. Schwartz later told the court that he had used ChatGPT to supplement his legal research, that he had not used it before for that purpose, and that he had not understood that it could invent material. He said he had asked the chatbot whether one of the cases was real, and it had replied that it was and could be found in standard legal research databases.

What the evidence shows

Avianca's lawyers told the court they could not locate many of the cited cases. Judge P. Kevin Castel then ordered Mata's lawyers to provide copies. They submitted what purported to be excerpts of the decisions, which had also been generated by ChatGPT. In a May 2023 order, the judge wrote that six of the submitted cases appeared to be bogus judicial decisions with bogus quotes and bogus internal citations. Some of the fake decisions named real judges as their authors and real airlines as parties, which made them seem more plausible.

At a hearing in June 2023, Schwartz apologized and said he had been duped by the technology. On June 22, 2023, Judge Castel found that the lawyers had acted in bad faith, particularly by continuing to stand by the fake cases after their existence was questioned. He fined Schwartz, LoDuca and their firm $5,000 and ordered them to send letters to their client and to each judge falsely identified as the author of a fake opinion. Separately, the court dismissed Mata's lawsuit as untimely.

Why it happened

Large language models such as ChatGPT generate text by predicting likely sequences of words based on patterns in their training data. They do not look up facts in a database unless they are specifically connected to one, and they do not have a built-in way to distinguish between real and invented information. When asked for legal precedents, a model can produce text that looks exactly like a real case citation, with a plausible name, court, volume and page number, because it has seen thousands of real citations. This tendency to produce confident, fluent and false information is commonly called hallucination.

The model's assurance that the cases were real was itself a hallucination. Asking a language model to verify its own output is not a reliable check.

Aftermath

Mata v. Avianca was only the first of many such cases. Since 2023, courts in the United States, Canada, the United Kingdom, Australia and elsewhere have identified hundreds of filings containing citations to nonexistent cases or fabricated quotations generated by AI tools. Researchers have compiled public databases tracking these incidents. Lawyers have been fined, referred to disciplinary authorities and, in some cases, removed from cases. Self-represented litigants have also filed AI-generated briefs with fake authorities. Many courts now require lawyers to certify whether they used AI and to confirm that every citation has been checked.

The problem is not limited to law. AI tools have been documented inventing academic references, news articles, quotations and biographical details about real people.

Lessons for readers

Generative AI can produce fluent, confident and entirely false information. Its output should be treated as a draft, not as a source.
Verify every citation, quotation and statistic against an original source.
Do not ask an AI tool to verify its own claims. It may simply invent confirmation.
Plausible detail is not evidence of truth. Fabricated material often looks more polished than real material.
Professional responsibility does not transfer to software. The person who submits work is responsible for its accuracy.

Sources

Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), US District Court for the Southern District of New York, Opinion and Order on Sanctions, June 22, 2023.
Order to show cause, Mata v. Avianca, Inc., May 4, 2023.
Benjamin Weiser, "Here's What Happens When Your Lawyer Uses ChatGPT," The New York Times, May 27, 2023.
Damien Charlotin, AI Hallucination Cases database (ongoing).