Wall Street Law Firm Apologises For AI Errors

Why It Matters

The episode underscores the legal sector’s growing exposure to AI hallucinations, prompting tighter compliance and risk‑management measures. Courts are increasingly sanctioning such errors, threatening reputational and financial consequences for firms.

Key Takeaways

• Sullivan & Cromwell filed AI‑generated citations that didn’t exist
• Partner Andrew Dietderich apologized to a federal judge for the error
• Errors were spotted by opposing firm Boies Schiller Flexner
• Incident highlights gaps in firm’s AI review and compliance policies
• Courts increasingly sanction lawyers for AI hallucinations since 2023

Pulse Analysis

Artificial intelligence has become a staple in modern law firms, promising faster research, draft automation, and cost savings. Yet the same models that generate citations in seconds also produce "hallucinations"—fabricated case names, misquoted authorities, or entirely fictitious sources. Since 2023, scholars have documented a steady rise in court filings containing such errors, prompting judges to issue sanctions and warning letters. The legal community now wrestles with balancing efficiency gains against the risk of undermining the credibility of judicial documents.

Sullivan & Cromwell, a 140‑year‑old firm with over 1,000 attorneys, fell short of its own AI safeguards when a partner’s filing included fabricated citations. Andrew Dietderich, co‑head of global finance and restructuring, sent a letter of apology to the presiding federal judge after Boies Schiller Flexner flagged the inaccuracies during a bankruptcy proceeding involving Prince Global Holdings. The firm acknowledged that its internal review process failed to catch the hallucinations, despite policies designed to prevent AI‑generated misinformation. The public apology underscores the reputational stakes for top‑tier firms that rely on emerging technology without rigorous oversight.

The incident adds pressure on the legal industry to formalize AI governance. Courts are signaling zero tolerance for fabricated content, and bar associations are drafting guidance on model‑driven research tools. Firms are expected to implement layered checks: automated citation verification, human editorial review, and clear accountability logs. Early adopters who invest in robust validation frameworks can preserve client trust while still harvesting AI’s productivity gains. As regulators contemplate mandatory disclosure of AI‑assisted drafting, the Sullivan & Cromwell episode serves as a cautionary benchmark for every practice seeking to modernize responsibly.