AI Hallucinations in Court The Sullivan & Cromwell Lesson

Recorded as a fireside chat at the ABA Law Practice Division Spring Meeting in San Diego, this special episode examines reports that Sullivan & Cromwell filed a brief containing nonexistent or inaccurate case citations and later apologized to opposing counsel and the court.Steve Embry, Jennifer Ellis, and Michael Eisenberg discuss why AI hallucinations continue to appear in legal filings despite widespread awareness of the risks. They point to ignored firm policies, time pressure, overreliance on confident-sounding AI outputs, and a lack of understanding of generative AI's limitations as key contributing factors.The discussion also explores how these failures relate to lawyers' ethical duties of competence, candor, and supervision. The speakers emphasize that legal citations should always be independently verified using traditional legal research tools—not AI—such as Shepardizing in Lexis or Westlaw. They also stress that proper training and experience are essential for recognizing unreliable AI-generated results.00:00 – AI Hallucinations Hit BigLaw00:37 – Show Intro and Disclaimer01:16 – Sullivan & Cromwell Brief Fallout02:19 – Why Lawyers Keep Getting Burned04:19 – Competence Rules and Citation Checks06:53 – Workflow Reality Check08:55 – Training to Spot Hallucinations10:45 – Verifying Citations Without AI Tools12:55 – Closing Thoughts and Wrap-Up13:02 – Podcast Outro and Resources