United States v. Cohen
U.S. District Court, Southern District of New York · S.D.N.Y. · New York bar guidance
Conduct
Attorney filed pro-forma motion containing 3 fabricated citations Cohen had generated via Google Bard and passed to counsel.
Consequence
Rule 11 sanctions declined; conduct negligent but not in bad faith; opinion articulates attorney candor duty under generative AI.
Lesson
Early canonical AI-citation matter: Bard-named, attorney-of-record warning rather than sanction, March 2024 timestamp.
Verified May 8, 2026
- Citation
- United States v. Cohen, 18-CR-602 (JMF), 2024 WL 1193604 (S.D.N.Y. Mar. 20, 2024) (Furman, J.)
- Decided
- March 20, 2024
Summary
In a high-profile early AI-citation matter, attorney David M. Schwartz filed a motion on behalf of Michael D. Cohen seeking early termination of supervised release. The motion contained three nonexistent citations that Cohen had personally generated using Google Bard and passed to Schwartz as research suggestions. One citation Schwartz's filing presented as "United States v. Figueroa-Florez, 64 F.4th 223 (2d Cir. 2022)" did not exist as described; per Judge Furman, "64 F.4th 223 refers to a page in the middle of a Fourth Circuit decision that has nothing to do with supervised release." Schwartz did not independently verify the citations before filing. After the government flagged the fabricated citations, District Judge Jesse M. Furman issued an Order to Show Cause on December 12, 2023 directing Schwartz and his co-counsel E. Danya Perry to explain the conduct and address whether Rule 11 sanctions or other discipline were warranted.
- AI tool:
- Google Bard (specifically named; Cohen described it on the record as 'a super-charged search engine')
What sanction did the court impose?
Judge Furman declined to impose Rule 11 sanctions or other formal discipline in the March 20, 2024 opinion. The court credited Schwartz's testimony that he believed the cases had come from co-counsel and concluded the Rule 11 bad-faith standard had not been met, finding the conduct negligent but not in bad faith. The court used the opinion to articulate the duty of attorney candor in light of generative AI. The matter generated extensive contemporaneous press coverage and is now one of the canonical early entries in the AI-hallucination sanctions corpus alongside Mata v. Avianca and Park v. Kim.
Why does United States v. Cohen matter for law firms using AI?
Judge Jesse M. Furman of the Southern District of New York decided Cohen on March 20, 2024. The underlying motion was routine: a request for early termination of Michael D. Cohen’s supervised release, filed by attorney David M. Schwartz with E. Danya Perry as co-counsel. Three of the cited authorities did not exist. Cohen had personally drafted research suggestions using Google Bard and passed them to Schwartz, who filed without verification. Cohen later told the court that he had treated Bard as “a super-charged search engine” and did not understand the tool would produce fabricated citations. The citation presented as “United States v. Figueroa-Florez, 64 F.4th 223 (2d Cir. 2022)” did not exist as described; per Judge Furman, “64 F.4th 223 refers to a page in the middle of a Fourth Circuit decision that has nothing to do with supervised release.” The other two had no real counterparts at all.
Judge Furman split the analysis along the signature line. A non-lawyer client supplying suggested research is not sanctionable; the lawyer who signs and files without verifying is the Rule 11 subject. Schwartz was negligent, but the bad-faith standard was not met, so Rule 11 sanctions were declined. The opinion is now cited two ways: defense counsel rely on it for “no bad faith, no Rule 11,” and movants rely on it for the proposition that any client-supplied research is the lawyer’s verification problem the moment counsel signs.
Implications for your firm
Operational steps a firm reading this case may wish to consider documenting. Strategic and rule-application calls belong to your firm's attorneys.
- Cohen is the canonical citation for the proposition that an attorney's reliance on a client's AI-generated research without independent verification is negligent, but does not necessarily meet the Rule 11 bad-faith standard. Useful authority for sanctions briefing both ways: defense counsel cite Cohen for 'no bad faith → no Rule 11'; movant cite Cohen for the duty-to-verify principle.
- The case names a specific tool (Google Bard) on the record. Bard's branding has since been replaced by Gemini; Cohen is one of the few federal sanctions opinions that names Bard specifically, useful for any historical timeline of which AI tools have produced sanctioned hallucinations.
- The opinion's reasoning establishes that an attorney's conduct in filing AI-generated research from a non-attorney source (here, the client) is the attorney's own Rule 11 conduct, not the client's. Firms should treat any client-supplied 'research' as raw material requiring independent verification, with the lawyer's signature carrying the Rule 11 risk.