USPTO Disciplines Attorney for AI Hallucinations in Patent Record

USPTO breaks new ground in AI enforcement

The U.S. Patent and Trademark Office has issued its first disciplinary order explicitly predicated on generative AI use, sanctioning a California patent attorney for submitting fabricated citations to a patent's intrinsic record. The August 27 order in In re Brian E. Mitchell, Proceeding No. D2026-16, marks a significant expansion of AI-related enforcement beyond the case law hallucinations that have dominated headlines.

According to IPWatchdog, which first reported the details, Mitchell used one AI tool to draft claim construction language and a second to review it while representing a patent owner in federal litigation. The AI-generated portions cited the patent's specification, figures, and prosecution history—but many citations pointed to portions of the record that did not exist or were inaccurately attributed.

Mitchell discovered additional errors himself and circulated a corrected chart the following day. The district court imposed no sanctions, the client suffered no prejudice, and the case later settled. The USPTO's Office of Enrollment and Discipline sanctioned him anyway.

Four violations, one new warning

The settlement agreement found violations of four rules: competence, diligence, misrepresentation, and conduct prejudicial to the administration of justice. The competence finding is particularly instructive—OED cited Mitchell's failure to understand AI research tool risks, inadequate review of citations to the intrinsic record, and insufficient inquiry before filing.

Notably, Mitchell had used a second AI tool specifically to review the first tool's output. That verification step did not satisfy his professional obligations, a finding that should concern practitioners treating AI review layers as sufficient safeguards.

The misrepresentation finding contained no requirement of knowledge, intent to deceive, or scienter. Despite prompt correction, full cooperation, and the absence of court sanctions, OED found the violation established.

Why it matters

The order's published notice extends a pointed warning: AI citation errors are not confined to external sources like case law but "may extend to the intrinsic evidence of patent and trademark applications and their file wrappers." This represents a genuine shift in enforcement focus. The verification workflows the profession has built over three years concentrate almost entirely on checking citations against Westlaw and Lexis. None of those tools catch a wrong column-and-line cite to a specification, a nonexistent figure number, or a prosecution history quotation that was never written.

The order also clarifies jurisdictional reach. Mitchell has never prosecuted a patent application or appeared before the PTAB or TTAB—his registration number is functionally dormant. The conduct occurred entirely in federal district court. Patent litigators holding registration numbers they rarely use should note: the number creates ongoing obligations regardless of practice area.

Matos wasn't actually about AI

The order corrects a widespread misunderstanding about In re Anthony Matos, the March 2025 discipline case routinely described as the USPTO's first AI enforcement action. The stipulated facts in Matos record that while he used AI to learn about TTAB procedures, he did not use it to draft the brief containing fabricated citations. His violations stemmed from failing to read cases he found through internet searches—not from model hallucinations. The AI-focused CLE requirement in that order created its reputation, but the predicate conduct was different.

Mitchell's case is the first where generative AI use is the explicit basis for the violations. The distinction between the two cases also explains the different sanctions: Matos received a reprimand plus mandatory CLE; Mitchell received only a reprimand. The difference turned on response time—Mitchell corrected errors the next day, while Matos let nearly two months pass after being put on notice.

Details of the disciplinary proceeding were first reported by IPWatchdog.