The lawyer who cited six court cases that did not exist
· 4 min read
A passenger named Roberto Mata sued the airline Avianca, claiming he had been injured by a serving cart on a flight. Ordinary case, ordinary filing. Then his lawyers filed a brief opposing the airline's motion to dismiss, and cited six previous decisions as precedent.
None of the six existed. Every one had been produced by ChatGPT, complete with case numbers, judges' names and quoted passages, and every one was invented.
How it happened
Steven A. Schwartz, an attorney at Levidow, Levidow & Oberman who had been licensed in New York for over thirty years, had used ChatGPT for legal research. He stated in an affidavit that he had never used it as a research source before and was unaware its output could be false.
The detail that turned an error into sanctions is what came next. When the court questioned whether the cases were real, the filings stood by them. Judge P. Kevin Castel wrote that the respondents had abandoned their responsibilities when they submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question.
The court imposed a $5,000 sanction on Schwartz, his colleague Peter LoDuca and the firm. Reporting on the order also describes the lawyers being directed to notify the real judges whose names had been attached to the fabricated opinions.
Why a spell-check mentality fails here
The brief was well written. The citations were correctly formatted. Nothing about the document looked wrong, because a language model is optimised to produce text that looks right — and a fabricated citation looks exactly like a real one until somebody checks whether the case exists.
That is the whole shape of the problem. Hallucinations are not typos. They do not read as errors. They read as facts, which is why they survive proofreading and get filed.
What would have caught it
Every one of the six citations was a claim about a document that could be looked up. A check that searches for each named case and reports which ones no source mentions catches all six before the brief leaves the building.
The signal is a specific one: a subject that is otherwise well documented, with this particular assertion appearing nowhere. That is different from “we could not find it”, and it is the strongest indication that something was invented rather than merely obscure.
Paste a draft and see what holds up — a check a day, free, no card.
Sources
CNN: Lawyer apologizes for fake court citations from ChatGPT · Association of Corporate Counsel: Practical lessons from Mata v. Avianca
Claim Check reads a draft, pulls out every factual claim, and checks each one against a real source before you publish. Try it on something you are about to publish.