Sixth Circuit Signals Zero Tolerance for AI Hallucinations in Appellate Practice

In United States v. John Farris, No. 25-5623 (6th Cir. April 3, 2026), the U.S. Court of Appeals for the Sixth Circuit issued a cautionary opinion that every practicing attorney—particularly those experimenting with artificial intelligence—should read closely. The appeal was taken from the United States District Court, Eastern District of Kentucky at Covington, and involved Farris’s challenge to his sentence of imprisonment. Notably, the court did not reach the merits of the defendant’s sentencing appeal. Instead, it focused exclusively on serious professional lapses by appointed counsel arising from the misuse of AI in drafting appellate briefs.

The appointed attorney admitted that he used Westlaw’s CoCounsel AI tool to generate drafts of both his principal and reply briefs. However, he failed to adequately verify the output before filing. The result was deeply problematic: the briefs included fabricated quotations attributed to real cases and misstatements of binding precedent. Although the cited cases existed, the quoted language did not—an increasingly common form of AI “hallucination.”

The Sixth Circuit emphasized that this distinction offers no safe harbor. Whether an attorney cites a nonexistent case or inaccurately quotes a real one, the ethical violation is the same: misrepresentation to the court. The opinion underscores that technological convenience does not diminish core professional duties—particularly the duties of competence, candor, and supervision.

Importantly, the court highlighted that attorneys cannot delegate verification responsibilities to non-lawyer staff or rely blindly on AI outputs. Even when using reputable, industry-backed tools, lawyers must independently confirm the accuracy of every citation and legal proposition. The court framed this obligation as an extension of longstanding ethical rules, not a new standard created by emerging technology.

The consequences for counsel were significant. The court denied compensation under the Criminal Justice Act, referred the matter for potential disciplinary proceedings, removed counsel from the case, and ordered the appointment of new appellate counsel—delaying resolution of the defendant’s appeal. The decision makes clear that AI-related errors are treated no differently than traditional forms of professional misconduct.

Questions? Contact Jared Burke for more information.