The Enterprise AI Failure Casebook

Lawyers filed court opinions that ChatGPT made up, then stood by them

A New York firm cited judicial decisions that did not exist, and a federal judge sanctioned two of its lawyers and the firm.

Levidow, Levidow & Oberman P.C. · June 22, 2023 · Updated October 31, 2026

On March 1, 2023, Peter LoDuca of the law firm Levidow, Levidow & Oberman filed a sworn opposition to a motion to dismiss in the federal court for the Southern District of New York. His client, Roberto Mata, was suing Avianca. Mata said a metal serving cart had struck his left knee on a flight from El Salvador to John F. Kennedy Airport. Avianca argued the claim was time-barred. The opposition answered with decisions it said were published in the Federal Reporter, the Federal Supplement and Westlaw.

LoDuca signed it, but he had not written it. His colleague Steven A. Schwartz did the research and the writing. LoDuca's name was on the case because Schwartz was not admitted to practice in that court. LoDuca read the draft for style and checked none of the cases it cited. The firm used a research service called Fastcase and had no Westlaw or LexisNexis account. Its Fastcase plan had limited access to federal cases. So Schwartz tried ChatGPT, which he had never used. He testified that he "falsely assumed" it was "like a super search engine." He asked it to "provide case law" and to "show me more cases." In the court's words, "the chatbot complied by making them up." He did not have the full text of any of them. He quoted the chatbot's excerpts.

On March 15, Avianca's lawyers told the court they could not locate most of the cases. Nobody at the firm withdrew the filing. LoDuca forwarded the reply to Schwartz without reading it. The court ran its own search and could not find several of the cases. On April 11 and 12, Judge Castel ordered LoDuca to file copies. LoDuca asked for more time, writing that he was out of the office on vacation. The court later found that statement was false. Schwartz was the one away.

On April 25, LoDuca swore to an affidavit that attached excerpts of all but one of the decisions. Schwartz had drafted it and walked it into LoDuca's office, where LoDuca signed it. One attachment, "Varghese v. China Southern Airlines," was presented as an Eleventh Circuit opinion. The clerk of that court confirmed it was not one of its rulings. The judge called its legal analysis "gibberish." Its quotation marks were often unpaired, and it ended without a conclusion.

On May 4 the court ordered LoDuca to show cause why he should not be sanctioned. On May 25 Schwartz admitted, in the first such statement from any of them, that the cases came from ChatGPT. The lawyers later acknowledged that six of the decisions were generated by ChatGPT and did not exist. His affidavit said ChatGPT had been used to "supplement" his research. It attached screenshots in which he asked ChatGPT whether "Varghese" was real and whether the other cases were fake. ChatGPT answered that its cases were real and could be found on Westlaw, LexisNexis and in the Federal Reporter. The court found his accounts of when he asked those questions shifting and contradictory. At a hearing on June 8 the judge asked, "It was your research, correct?" Schwartz answered, "Correct." He also testified, "I just never thought it could be made up."

On June 22, 2023, the court found that both lawyers had acted in bad faith, through conscious avoidance and false and misleading statements, and held the firm jointly responsible. It ordered them to pay a penalty of $5,000 into the court's registry, jointly and severally. It ordered them to send letters, with the opinion and the hearing transcript, to Mata and to each real judge falsely named as the author of a fake opinion. It did not require an apology.

What failed

The court's opinion opens by saying there is nothing inherently improper about using a reliable AI tool for assistance, and that the rules make lawyers the gatekeepers of the accuracy of what they file. It sanctioned the lawyers for submitting fake opinions and standing by them after the court's orders called their existence into question. By his own account, one lawyer turned to a chatbot because the firm's Fastcase account had limited access to federal cases, and he testified that he assumed it was a search engine. Its output went to a court as law. The lawyer who signed under penalty of perjury checked none of it. When the other side said it could not find the cases, and when the court ordered copies, nobody told the court the truth until May 25. The firm's remedies came after the fact: a larger Fastcase subscription and a mandatory legal education program on technological competence and artificial intelligence.

Sources

1. U.S. District Court for the Southern District of New York, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Opinion and Order on Sanctions (ECF 54, June 22, 2023), pp. 1-2; Findings of Fact ¶¶ 1-3, 5-11, 13-14, 16-18, 20-22, 25-27, 36-37, 39-42, 44-47, 50-51 (pp. 3-21); Conclusions of Law ¶¶ 21, 23-25, 30, 33 (pp. 29-33); Conclusion (p. 34); Appendix B (pp. 41-43)