Lawyers sanctioned for filing ChatGPT-fabricated case citations

Judge P. Kevin Castel of the US District Court for the Southern District of New York sanctioned two attorneys, Steven Schwartz and Peter LoDuca, and their firm Levidow, Levidow & Oberman $5,000 for filing a legal brief containing citations to court decisions that did not exist. The case, a personal-injury suit against Avianca airline, became the first widely reported instance of AI-fabricated legal citations reaching a court sanction.

Schwartz had used ChatGPT to research supporting case law and, when he asked the chatbot to confirm the cases were genuine, it assured him they could be found in Westlaw and LexisNexis. They could not: ChatGPT had invented six decisions complete with plausible-sounding names, docket numbers and quoted reasoning. Opposing counsel and then the court were unable to locate any of them, and when Castel ordered the plaintiff’s team to produce the cases, Schwartz submitted copies that were themselves fabricated by the chatbot.

Castel’s order described the submitted opinions as containing “gibberish” legal reasoning and found the attorneys’ conduct met the standard for bad faith, since Schwartz had continued to vouch for the citations even after doubts were raised rather than checking a primary source. The case did not turn on any special AI-specific rule; the court applied the ordinary standard that a lawyer is responsible for verifying what they file, regardless of how it was drafted.

The episode became a reference point rather than an isolated embarrassment. Hallucinated citations recurred in filings through the following years — an insurance case sanctioned two firms $31,100 in May 2025 is one of many — and bar associations subsequently issued guidance treating unverified AI-generated legal research as a professional-conduct risk rather than a novelty.