Should AI briefs fail review, sanctions remain possible

On September twenty-four, a court in Clark v. Social Security, Commissioner of ordered Rich to show cause why she should not be sanctioned, including through a fine, after a brief prepared with generative artificial intelligence contained factitious citations. CourtListener RECAP dockets: generative AI and sanctions or hallucinations reports that the filing says Rich used generative artificial intelligence in preparing the plaintiff’s brief, and that the court identified the factitious citations before issuing the show-cause order. The filing also says Rich responded to that order and used AI in preparing the brief. The immediate news is therefore not that a court has imposed a final penalty. It is that AI-assisted legal work has entered the ordinary accountability process for filed briefs, with a fine expressly on the table.

That is a direct example of why AI does not eliminate the consequences of bad legal work. A brief is finished legal work only if its authorities and factual support survive review. The Clark order shows the cost of failure moving through the filing itself: a defective brief can trigger a demand for an explanation and possible sanctions even when AI was used as part of its preparation. The distinction matters for AI-native law firms. Their advantage cannot be reduced to producing text quickly. It depends on delivering work that remains accurate, reviewable and defensible after it reaches a court. CourtListener: AI and sanctions (opinions) reports that R. M. v. Mohr also concerns an order addressing artificial intelligence and fabricated material. That does not establish a general rule for every AI-assisted filing, but it shows that the consequences are attached to the legal work delivered, not removed by the tool used to produce it.

Clark proves one narrower point and no more. It establishes that one court began a sanctions process over one brief containing factitious citations after generative AI was used in its preparation. It does not establish that every use of AI in legal drafting produces defective work, that every AI-assisted brief warrants a sanction, or that an AI-native firm has failed. The order is a show-cause step, not a final sanction, so the stronger reading would require a later order imposing a penalty and explaining that the AI-assisted drafting and the defective citations justified it. A different filing shows why verification is the dividing line rather than AI use alone: in Macias v. CCADC, CourtListener RECAP dockets: generative AI and sanctions or hallucinations reports that OpenAI Codex assisted with the document and that the filer certified personal review of the AI-assisted portions, including all citations. That certification does not prove the filing is correct, but it identifies the control that the Clark matter says was missing in practice.

The next fact to watch is the court’s order after Rich’s response to the show-cause directive. A sanction or fine, accompanied by findings about the factitious citations and the preparation of the brief, would strengthen the claim that AI-assisted legal work remains subject to ordinary professional consequences. An order declining to impose sanctions would narrow the lesson without erasing the underlying failure: the court still required an explanation for the citations. We should also watch whether later filings use the kind of personal-review certification described in Macias and whether those filings survive scrutiny. For AI-native law, the relevant output is not a promise that the system can draft. It is a filing that can withstand the court’s next order.