Wall Street law firm Sullivan & Cromwell apologized to a federal judge after filing a court submission that contained roughly 40 incorrect citations and other errors traced to AI hallucinations. Andrew Dietderich, co-head of the firm’s global restructuring group, wrote to Chief Judge Martin Glenn of the US Bankruptcy Court for the Southern District of New York that the firm “deeply regret[s]” the mistakes and that he accepts responsibility for the failure to ensure accuracy under Local Bankruptcy Rule 9011-1(d).
Dietderich said Sullivan & Cromwell has policies governing the use of AI tools that include a citation-review process, but those procedures were not followed in this instance. “Regrettably, this review process did not identify the inaccurate citations generated by AI, nor did it identify other errors that appear to have resulted in whole or in part from manual error,” he wrote. The errors appeared in an emergency motion submitted nine days earlier.
The episode underscores the risks posed by AI-assisted drafting in high-stakes legal work when oversight lapses. Legal technologist Damien Charlotin’s database has logged 1,334 incidents of AI hallucinations in court filings globally, with more than 900 in the US; most reported instances involve fabricated citations, though some include AI-generated legal arguments.
Sullivan & Cromwell said it took immediate remedial steps and launched a full review of how the errors occurred. The firm is evaluating whether additional training or enhancements to its internal review processes are warranted. Dietderich also noted that the inaccuracies were brought to the firm’s attention by rival counsel; he called Boies Schiller Flexner LLP to thank them for flagging the issues and to apologize.
Sullivan & Cromwell is among the largest US law firms by revenue, ranking 30th on the AmLaw Global 200, and has acted in major matters including the bankruptcy of crypto exchange FTX. The firm said it is conducting an internal investigation and will consider further measures to prevent similar AI-related errors in future filings.
