United States v. Michel
U.S. District Court for the District of Columbia · D.D.C. · District of Columbia bar guidance
Verified April 26, 2026
- Citation
- United States v. Michel, No. 1:19-cr-00148 (D.D.C. Aug. 30, 2024) (Kollar-Kotelly, J.)
- Decided
- August 30, 2024
Summary
Pras Michel, convicted in April 2023 on ten counts including conduit campaign contributions, witness tampering, and unregistered foreign lobbying, moved for a new trial under Fed. R. Crim. P. 33. Among 14 asserted errors, Michel argued that defense counsel David Kenner had relied on the experimental generative AI tool EyeLevel.AI to draft his closing argument, which misattributed a Puff Daddy lyric to the Fugees. Judge Colleen Kollar-Kotelly accepted that some errors occurred (including ineffective assistance) but held the AI-driven misattribution did not prejudice the verdict given the strength of the government's evidence.
- AI tool:
- EyeLevel.AI (generative AI)
- Sanction amount:
- None
What sanction did the court impose?
Motion for new trial denied. Court held Michel did not show a reasonable probability the result would have differed had Kenner correctly attributed the lyric, and that the cumulative errors did not warrant reversal.
Why does United States v. Michel matter for law firms using AI?
Michel is the highest-profile generative-AI-in-litigation case to date and an instructive counterexample to the sanctions narrative. No fabricated citations were filed; instead, defense counsel used EyeLevel.AI to help draft a closing argument that contained a factual misattribution (a Puff Daddy lyric attributed to the Fugees). The court treated the AI use as part of an ineffective-assistance analysis rather than a Rule 11 problem, and ultimately held that the misattribution, even combined with other errors, did not move the needle against the government’s evidence. For managing partners, the lesson is that AI-related missteps in litigation can survive a new-trial motion when the underlying record is strong, but the reputational exposure of a closing-argument error blamed on AI is a separate, and substantial, cost.
Sources
Primary sources
- Government primary source
- CourtListener docket (primary source)
- CourtListener opinion (primary source)
- Full text of the August 30, 2024 memorandum opinion was not parseable directly via WebFetch (PDF binary); ruling details corroborated through Reason/Volokh Conspiracy summary that quotes the opinion and the CourtListener docket entry.