At a glance
- Company
- Levidow, Levidow & Oberman P.C.Law firm
- Size
- Not stated in the source The court record names one equity partner.
- Industry
- Professional services
- Where and when
- United States (federal court in New York). March–June 2023.
- Who produced the figures
- Court or tribunal recordEverything on this page comes from the court’s published opinion and order on sanctions.
What they did
The firm acted for a passenger suing the airline Avianca. One of its lawyers needed federal case law that the firm’s research service did not fully cover. He turned to ChatGPT, which he had not used before. He told the court he “falsely assumed” it was “like a super search engine”.
ChatGPT gave him summaries of court decisions that supported his argument. The firm filed them on 1 March 2023.
The numbers
- Six of the cited decisions did not exist. ChatGPT had made them up, with fake quotes and fake citations.
- About twelve weeks passed between the filing and the lawyers telling the court what had happened. The airline’s lawyers questioned the cases on 15 March. The court ordered copies in April. The lawyers then filed what they presented as copies or excerpts of the decisions. Those were fake too.
- $5,000 was the penalty, imposed jointly on the firm and two lawyers on 22 June 2023.
- The court also ordered them to send its opinion to their client and to each judge named as the author of a fake decision.
What happened next
The firm told the court it had widened its research subscription. It also arranged training for its lawyers on technology and AI tools. The court took this into account.
What the case shows
- The court did not ban AI. The judge wrote that there is “nothing inherently improper about using a reliable artificial intelligence tool for assistance”.
- The duty to check stayed with the lawyers. The judge wrote that existing rules give lawyers “a gatekeeping role” to make sure their filings are accurate.
- The user did not know what the tool was. He thought it was a search engine. A short test of what staff know about AI would have caught this. A self-rating would not.
- Standing by the output made it worse. The court said the outcome would have been “quite different” if the lawyers had admitted the problem when first asked.
What this says about the practices
Practices that were missing
- Test AI knowledge instead of asking staff to rate their own The lawyer had not used ChatGPT before and assumed it was “like a super search engine”. Do next · Effort: medium (weeks, a few people)
- Budget for the work AI creates The made-up decisions were filed without being checked. The court said the duty to check stayed with the lawyers. Do first · Effort: low (days, one person)
- Keep a person checking anything that goes to a customer or carries legal weight Nobody looked the six decisions up before they were filed. The court said lawyers keep “a gatekeeping role” over what they file. Do first · Effort: medium (weeks, a few people)
- When AI output goes wrong, correct it fast and in the open The lawyers stood by the fake cases for about twelve weeks. The court said the outcome would have been “quite different” had they admitted the problem at once. Do next · Effort: low (days, one person)
What the company changed afterwards
Sources
- Mata v. Avianca, Inc., No. 22-cv-1461 (PKC). Opinion and Order on Sanctions, 22 June 2023. US District Court, Southern District of New York. storage.courtlistener.com