AI briefs face sanctions if citations are factitious

A court ordered Rich to show cause on September 24, 2026, as to why she should not be sanctioned, including with a fine, after her brief in Clark v. Social Security, Commissioner of included factitious citations. The CourtListener RECAP docket says Rich utilized generative artificial intelligence in preparing the brief and responded to the show-cause order. The order makes the immediate event concrete: AI use appeared in a filing, the filing contained false citations, and the court moved toward possible sanctions.

The consequence matters because an AI-assisted filing is still judged as legal work submitted to a court. The court did not need to resolve whether generative AI is useful in general before requiring Rich to answer for the citations in this brief. That is the practical limit on any workflow that treats generation as completion: the filing remains exposed to review, and defects can trigger a formal response from the court. The point is not that every use of AI produces bad work. It is that useful output does not erase responsibility for the submitted result. The CourtListener: AI and sanctions opinions account describes Douglas v. Deutsche Bank National Trust Co. as a cautionary tale about misuse of artificial intelligence and its consequences for attorneys, while the Clark docket supplies a specific pending sanction inquiry.

Clark establishes that one AI-assisted brief contained factitious citations and that the court ordered a show-cause response that could include a fine. It does not establish that AI-assisted legal work is generally unreliable, that every AI use warrants sanctions, or that an AI-native legal firm has failed. The order concerns Rich and the brief identified in the docket; it says nothing about a firm’s product, review system, pricing model, or ability to deliver finished legal work. A stronger reading would require a completed sanction order, findings about how the citations entered the filing, and evidence connecting those findings to a repeatable legal-work process rather than to one filing. Another CourtListener RECAP docket, Macias v. CCADC, describes a filing that disclosed OpenAI Codex assistance and certified personal checking of the AI-assisted portions, including citations. That filing shows a different control described on the face of a document, but it does not establish that the checking was effective or that the court accepted the work. The relevant standard remains the submitted legal product, not the presence or absence of an AI label.

The next observable fact is the court’s order on the Clark show-cause response. A sanction or fine would confirm that factitious citations in an AI-assisted brief can produce a concrete judicial penalty in this matter. An order declining sanctions would narrow the lesson, but it would not erase the show-cause order or the factual problem identified in the brief. We will treat the final order, and any findings about review of the citations, as the test of whether the consequence reaches beyond procedural scrutiny to an actual sanction.