Overview
No lawyer wants to explain to a client that they were thrown off a case for relying on hallucinated citations. But that is what happened recently in Withers v. City of Aberdeen, a fee dispute between a lawyer and a city. Both sides filed briefs with citations that generative AI had invented. In a June 8 order, Judge Sharion Aycock removed all four lawyers from the case and barred the two out-of-state lawyers from appearing in any case in the Northern District of Mississippi for two years. Their “practice of blindly relying on technology,” she wrote, “resulted in the hallucinatory citations contained in their respective filings.”
Five days earlier, on June 3, the Ninth Circuit made the same point in a precedential order. In LNU v. Blanche, the court suspended two immigration lawyers after briefs in their case cited nonexistent cases and fabricated quotations, errors later defended as mere typos. The panel had no patience for the typo defense. “A competent and diligent attorney,” it wrote, “must do more than prompt generative AI, check that the citations provided by the AI are real and the subject matter roughly on point, and call it a day. A competent and diligent attorney must also read and reason.”
The panel saved its harshest words for the lawyers’ lack of candor. It wrote that there is “no upside to denying the use of generative AI or to passing off an AI hallucination as an innocent typographical error.” Because the order carries precedential force, lawyers practicing in the Ninth Circuit now have circuit-level guidance to reckon with.
Those sanctions fit a larger pattern. A database kept by the researcher Damien Charlotin now logs more than 1,600 cases worldwide (more than 1,100 in U.S. courts) in which courts have flagged AI hallucinations, usually fabricated citations, up sharply from a year ago and still climbing. Judges have warned the bar since the first fake-case scandals of 2023. For a while, the usual price was a modest fine and an embarrassing order. That did not slow the spread. Courts are now reaching for sanctions that actually sting.
The duty these sanctions enforce is not new. In the district courts, Rule 11 has long made a lawyer’s signature on a court filing a certification that the filing rests on a reasonable inquiry, and the appellate courts apply the same duty through their own rules. That responsibility does not pass to the model that drafted the brief.
Generative AI tools are very good at work that humans often find difficult or boring, such as analyzing prior art, file histories, technical specifications, and case law. The problem is that these tools can make mistakes no human would make. As pertinent here, generative AI is notorious for confidently quoting documents that do not exist. Whether or not you feel confident in your tool, courts are going to insist that lawyers read the underlying sources they cite and confirm their representations about them. The signature on the brief is a person’s, and that person answers for it.
The Weekly Inference is a recurring Step Into IP feature. Nothing here is legal advice.
