Sullivan & Cromwell has admitted to making significant filing errors caused by AI hallucinations in a letter to a US bankruptcy court.
The letter, which was sent by the co-founder of the firm’s global bankruptcy group, Andrew Dietderich, to Chief Judge Martin Glenn in New York, disclosed that an emergency motion filed on 9 April contained several inaccurate citations and misquoted the US Bankruptcy Code.
Dietderich defined hallucinations as instances where AI tools “fabricate case citations, misquote authorities or generate non-existent legal sources”.
He wrote: “We deeply regret that this has occurred. The firm maintains comprehensive policies and training requirements governing the use of AI tools in legal work. These safeguards are designed to prevent exactly this situation. The firm’s policies on the use of AI were not followed in connection with the preparation of the motion.”
He conceded that the review process had failed to identify the inaccurate citations generated by AI use in this instance. Some 42 inaccuracies were included in the motion.
The hallucinations were flagged by Boies Schiller Flexner, the opposing counsel in the case, whom Dietderich thanked.
Sullivan & Cromwell has been acting as the British Virgin Islands court-appointed litigators in actions against Prince Group Holdings and its Chinese owner, Chen Zi, who has been accused of involvement in a cryptocurrency fraud scheme linked to forced labour.
Dietderich said the firm had reviewed all of the filings in the case and confirmed there were no additional AI-generated inaccuracies. It is also reviewing the circumstances that led to the errors to assess whether protocols need to be reviewed.
AI-hallucinated case citations have become relatively commonplace. A database created by Paris-based researcher and law lecturer Damian Charlotin has identified 1,334 such cases globally.
Last week, consultants Gartner advised general counsel to consider direct AI insurance coverage, to protect themselves from AI-related risks due to the acceleration of incidents, including hallucination episodes. This is because insurers are increasingly placing AI exclusion clauses into traditional policies.
Email your news and story ideas to: [email protected]
