Should AI tools fail to verify outputs, attorneys still face consequences

On September 24, the docket in Clark v. Social Security, Commissioner of says the court ordered Rich to show cause why she should not be sanctioned, including through a fine, for including factitious citations in a plaintiff’s brief. CourtListener RECAP dockets says the brief used generative artificial intelligence and that Rich responded to the show cause order by acknowledging AI use in preparing it. The docket does not report a final sanction or a final finding about the citations.

The event exemplifies the basic limit on AI-assisted legal work: a tool can help prepare a filing, but the filing remains exposed to the consequences of bad citations. AI does not eliminate the consequences of bad legal work. Here, the alleged defect is not abstract model performance. It is a court inquiry into factitious citations in an actual brief, with a fine expressly within the possible response. That is a direct connection between unverified output and legal accountability, even though the docket has not yet established that a fine will be imposed.

The order establishes that AI use can be part of a sanctions inquiry when a filing contains factitious citations. It does not establish that AI caused every defect in the brief, that Rich will ultimately be sanctioned, or that every use of AI in legal work creates the same result. It also says nothing about an AI-native firm as a business model. The stronger reading would require a final order identifying the conduct that warranted sanctions, followed by comparable orders showing that the result is not confined to this filing. At present, the event is one show-cause proceeding, not a general rule.

Other court records point to the verification step that the Clark proceeding puts under pressure. In Macias v. CCADC, CourtListener RECAP dockets says OpenAI Codex assisted in preparing a document and that the filing included a personal-review certification covering the accuracy of the AI-assisted portions and all citations. In Douglas v. Deutsche Bank National Trust Co., CourtListener: AI and sanctions (opinions) describes misuse of artificial intelligence and its consequences for attorneys. Neither record supplies a universal standard, but together they show the relevant distinction: assistance is not the same as review, and review is not the same as immunity from consequences.

The next fact to watch is a sanctions order in Clark v. Social Security, Commissioner of. A final order imposing a fine or another sanction, with findings tying the outcome to the factitious citations and the AI-assisted preparation, would strengthen the reading that AI use leaves ordinary legal accountability intact. A decision declining to sanction Rich, or finding that the citations were not established as a result of the AI use, would narrow it. More broadly, a later order that addresses whether attorneys must verify AI-generated citations would turn this proceeding from a warning about one brief into a clearer operating requirement. We will treat that order, not the existence of AI use alone, as the consequential fact.