analysisAI Hallucinations
Mata v Avianca: the case that started it all
Two New York attorneys, six ChatGPT-invented cases, and a $5,000 sanction — the June 2023 ruling that created the legal AI hallucination era.
In June 2023, a US federal judge fined two New York attorneys $5,000 for filing a brief containing six cases that did not exist. ChatGPT had invented them. Mata v Avianca was the first widely reported sanction for AI-fabricated citations — and the template for the hundreds of incidents that followed.
What happened
Roberto Mata sued the airline Avianca over a knee injury from a metal serving cart. When Avianca moved to dismiss, Mata’s lawyers filed an opposition citing cases like Varghese v China Southern Airlines — persuasive, on point, and entirely fictional.
Opposing counsel could not find the cases. Neither could the court. Asked to produce them, the lawyers went back to ChatGPT, which obligingly generated full fake opinions, complete with invented docket numbers and judicial signatures. They filed those too.
What the court decided
Judge P. Kevin Castel’s sanctions order of 22 June 2023 made two findings that still shape how courts respond today. Submitting fake cases was not itself the unforgivable act — the lawyers “abandoned their responsibilities” when they stood by the fake opinions after being challenged. And existing rules were sufficient: Rule 11 already required a reasonable inquiry into the law being cited. No new AI rule was needed to sanction the conduct.
The penalty was $5,000 jointly — modest by later standards. A tracked sanction in Oregon reached $110,000 within three years, and courts in England, Canada, Australia and South Africa have since referred lawyers to their regulators for the same pattern.
Why it still matters
Three years on, the Mata pattern repeats with striking fidelity: a lawyer under time pressure, a tool that produces confident fabrications, a failure to verify, and — the aggravating factor courts punish hardest — doubling down when challenged. The Divisional Court in England made the same point in Ayinde in 2025; the Supreme Court of British Columbia made it in Zhang v Chen in 2024.
The verification duty was always there. Mata was simply the first time a court applied it to generative AI — and the reason this tracker exists.