When AI made up court cases: the lawyers who got caught
Fabricated citations have been submitted to courts in over a thousand documented proceedings. Here are the cases that established the pattern, and what the sanctions now look like.
The first one is still the most instructive. In 2023, a personal injury claim against an airline produced the most-cited example of what happens when a lawyer trusts a chatbot’s research.
Mata v. Avianca (S.D.N.Y., 2023)
Roberto Mata sued Avianca over a knee injury from a serving cart. Opposing counsel could not find the cases his lawyers had cited. Neither could the judge.
Six of the decisions in the brief did not exist. They had been produced by ChatGPT, complete with docket numbers, internal quotations and, the detail that gives the case its teeth, attribution to real, sitting federal judges. A fabricated opinion called Varghese was presented as an Eleventh Circuit decision; Judge P. Kevin Castel described its reasoning as “gibberish.” Others were attributed to Judge Barrington D. Parker and Judge Reggie B. Walton, both real people who never wrote them.
What drew the sanction was not the initial mistake. It was what came after: when challenged, the lawyers filed copies of the fake opinions rather than checking whether they existed. Judge Castel found bad faith, citing “acts of conscious avoidance and false and misleading statements to the Court,” and imposed a penalty of $5,000, jointly and severally, on both attorneys and their firm.
Five thousand dollars, shared three ways, now looks like a bargain.
It did not stay an isolated incident
Damien Charlotin, a research fellow at HEC Paris, maintains a public database of court decisions in which AI-hallucinated material was identified. It passed 1,600 entries during 2026, spans dozens of jurisdictions, and its inclusion rule is strict: only proceedings where a court or tribunal actually found that a party relied on hallucinated content, not mere allegations.
The database is the single best answer to anyone claiming these are rare edge cases. It is a running tally, and it grows daily.
The sanctions got serious
Two recent examples show where courts have landed:
Couvrette v. Wisnovsky (D. Or., December 2025). Fifteen fabricated case citations and eight invented quotations across three briefs filed over five months. The court fined lead counsel $15,500 and awarded the defendants their fees on both the summary judgment and the sanctions briefing.
Whiting v. City of Athens (6th Cir., 2026). Briefs containing over two dozen fake citations and misrepresentations of fact. The court of appeals ordered the attorneys to reimburse the appellees’ full reasonable fees on appeal, pay double appellate costs, and each pay a further $15,000 in punitive sanctions into the court registry.
The trajectory from a shared $5,000 in 2023 to $15,000 per lawyer plus fees and double costs in 2026 is the story. Courts treated the first wave as an honest encounter with an unfamiliar tool. They have stopped.
Why legal citations are the perfect failure case
Case law is the ideal shape for a hallucination. Citations follow a rigid, learnable format. Party names are drawn from an ordinary distribution of surnames. Reporter volumes and page numbers are just plausible integers. A model can generate something with every surface property of a real citation without any of it being connected to a document that exists.
And unlike most model output, it is trivially checkable, which is why the failures surface in court records rather than staying private. The legal system is not unusually affected. It is unusually good at catching it.