In June 2023, New York lawyer Steven Schwartz stood before federal Judge P. Kevin Castel and tried to explain six court decisions that did not exist. He had asked ChatGPT to find authority for his client’s personal injury claim against the airline Avianca, and ChatGPT produced confident citations, complete with fabricated quotes and invented case numbers.
Schwartz’s co-counsel, Peter LoDuca, had filed the brief without checking the citations. Judge Castel sanctioned Schwartz, LoDuca, and their firm Levidow, Levidow and Oberman $5,000.
Why was Mata treated as a one-off?
At the time, the legal profession treated this as an isolated embarrassment, a cautionary tale that would prevent repetition. The fine was small. The facts were vivid. The lesson seemed obvious: do not file what you have not read.
Three years later, an independent academic database maintained by researcher Damien Charlotin at HEC Paris has documented more than 2,000 similar cases worldwide, with penalties escalating from that initial $5,000 to six-figure sanctions, bar suspensions, and cancelled trials.
What did the $5,000 fine actually start?
The Mata v Avianca fine was not the lesson. It was the first entry in a database that now needs daily updates. The same failure mode (an unverified model output filed as authority) has moved from a single New York courtroom to 44 jurisdictions. Check your governance exposure in three minutes.
This case is tracked in the Ecsper AI Risk Intelligence, which documents over 2,000 court cases involving AI-fabricated content.
