Sullivan & Cromwell Apologizes for AI Hallucinations in Court Filing

3 min readSources: Volokh Conspiracy, ABA Journal

Sullivan & Cromwell admitted using AI-generated, hallucinated legal citations in a 2026 court filing.

Why it matters: AI hallucinations in legal filings expose firms to professional, ethical, and financial risk. With law firms rapidly adopting generative AI, leaders must prioritize robust oversight to avoid sanctions and reputational harm.

  • Sullivan & Cromwell apologized to the U.S. Bankruptcy Court for using hallucinated citations in April 2026.
  • Cases of AI-generated fake citations in court filings surged to over 280 in 2025, a sevenfold increase.
  • Multiple law firms and attorneys have faced sanctions—including a $31,100 penalty and state bar discipline—for AI hallucinations.
  • Judges and bar associations have warned that lawyers must verify the accuracy of all filings, regardless of AI use.

Sullivan & Cromwell issued a formal apology to the U.S. Bankruptcy Court for the Southern District of New York in April 2026, after their filing was found to contain AI-generated hallucinated legal citations—references to non-existent cases or authorities. The incident echoes a series of similar errors across the legal industry as firms integrate generative AI tools into daily practice. (details)

  • Several firms—including Gordon Rees Scully Mansukhani LLP and K&L Gates—have faced public embarrassment, potential sanctions, and fines after submitting court documents riddled with AI-generated fake citations. (investigation)
  • Judges, such as Special Master Michael Wilner, have recounted nearly being persuaded by these hallucinated citations: “That’s scary.” (commentary)
  • By 2025, over 280 filings included hallucinated citations, and a global database tracked more than 490 such incidents by early 2026. (report)
  • Sanctions have become more common: in May 2025, Indiana and California courts imposed penalties of $6,000 and $31,100, respectively, on attorneys and firms submitting briefs laced with fictitious AI-generated citations. In June 2025, a Utah appellate attorney was fined $1,000. (coverage)

These episodes have prompted renewed calls from courts and bar associations to verify every citation, regardless of how the material is researched or drafted. The Utah appeals court emphasizes, “Every attorney has an ongoing duty to review and ensure the accuracy of their court filings.” (guidance)

In April 2026, the California State Bar opened disciplinary proceedings against three lawyers over AI hallucinations, underscoring the regulatory risks associated with failing to vet AI-generated work. (announcement)

The surge in these incidents highlights the vital need for human oversight, better training, and updated firm policies as legal AI adoption accelerates.

By the numbers:

  • 280+ — Court filings with AI hallucinated citations in 2025, a sevenfold rise over prior years
  • 490+ — Global instances of AI hallucinations in court cases tracked by early 2026
  • $31,100 — Sanction imposed on law firms in California for AI-generated fake citations in 2025

Yes, but: Details about the specific Sullivan & Cromwell case and the outcomes of some disciplinary actions remain undisclosed.

What's next: Bar associations and courts are expected to issue further AI guidance and possibly tighten professional conduct rules in 2026.