Sullivan & Cromwell Apologises for AI-Generated Court Errors
AI Errors Force Wall Street Law Firm's Court Apology
Sullivan & Cromwell LLP formally apologised to the United States Bankruptcy Court for the Southern District of New York in April 2026 after discovering that an emergency motion filed by the firm contained more than 40 inaccuracies generated by artificial intelligence MinterEllison.
The motion filed in the federal bankruptcy court included fabricated case citations, misquoted provisions of the Bankruptcy Code, and non-existent legal sources that had been generated by an AI tool used in the document's preparation. The errors ranged from invented precedents to material misstatements of statutory language, each capable of misleading the court on questions of law.
According to secondary reporting on the incident, Sullivan & Cromwell acknowledged the failures and took responsibility for submitting inaccurate materials to the judiciary. The firm's formal apology to the court marked a significant moment in the emerging debate over generative AI's role in legal practice, where the pressure to move quickly can collide with the profession's obligation to ensure accuracy and candor.
Broader Pattern of AI Errors in Legal Filings
The Sullivan & Cromwell incident is not isolated. Across U.S. federal courts, a growing pattern of AI-generated errors in legal briefs and motions has prompted increasing scrutiny from judges and bar associations Chosun Ilbo English. Courts have begun warning attorneys that reliance on generative AI without adequate human review constitutes a breach of professional responsibility.
The U.S. Courts Standing Committee has tracked these incidents as part of its broader review of technology governance in federal practice. While no specific monetary sanctions or disciplinary actions against individual attorneys have been reported in connection with the Sullivan & Cromwell filing, the case demonstrates how AI-assisted legal work can pose direct risks to the integrity of court proceedings.
What the Case Reveals
The April 2026 filing underscores a critical gap: AI language models can produce plausible-sounding but entirely fabricated legal authority—text that reads as coherent legal argument but contains no grounding in actual case law or statutory text. For courts and opposing counsel, detecting such errors requires line-by-line verification, imposing costs and delays.
Sullivan & Cromwell's swift acknowledgment and correction of the record prevented deeper prejudice to the parties before the court. The firm's response—taking responsibility rather than disputing the errors—established a precedent for how elite firms should handle AI-generated mistakes. However, the incident raises hard questions about when and whether AI tools should be used in filings where accuracy is non-negotiable, and whether current attorney review protocols are sufficient.
