AI-assisted drafting is finished legal work only if citations are verified

On September 24, the court in Clark v. Social Security, Commissioner of ordered Rich to show cause why she should not be sanctioned, including with a fine, after a brief prepared with generative artificial intelligence contained factitious citations, according to CourtListener RECAP. The docket says the brief used generative artificial intelligence and that the court issued the order after identifying the citations. The order makes the immediate issue concrete: AI-assisted drafting did not prevent the filing from drawing judicial scrutiny, and the court left sanctions on the table.

That is the relevant test for AI-native legal delivery. A system can produce a document quickly, but speed does not complete the legal task if the authorities in the document are false. The unfinished work is verification, and the consequence is not abstract: the court required a response to a possible sanction and specifically identified a fine. AI does not eliminate the consequences of bad legal work. For firms that deliver finished legal work rather than sell lawyer hours, that distinction matters more than the fact that an AI tool participated in drafting. The client receives the filing, while the court evaluates the filing and the process behind it. A product that cannot reliably distinguish usable authority from invented authority has not removed the accountability attached to legal work.

The order proves one narrower proposition: in one case, a court connected an AI-assisted brief containing factitious citations with a possible sanctions process. It does not establish that the court imposed a fine, that AI caused every defective citation, or that AI-assisted legal work generally produces bad results. It also says nothing about the quality of an AI-native firm as a whole. A separate CourtListener: AI and sanctions (opinions) item describes Douglas v. Deutsche Bank National Trust Co., Published Order as a cautionary tale about the misuse of artificial intelligence and its consequences for attorneys. That description supports the existence of judicial concern beyond Clark, but it is not a finding that every AI-assisted document is defective. The stronger reading would require a final sanctions ruling in Clark, findings about how the citations were generated and checked, and comparable orders showing that the problem persists across matters. Those facts are not present here.

The next observable fact is the court’s final ruling on the Clark show-cause order. It will show whether the court imposes a fine or another sanction, identifies a failure to verify the authorities, or declines to sanction after reviewing the response. That ruling will sharpen the distinction between AI use and defective legal work: the issue is not merely whether a tool was used, but whether the filing was checked before it reached the court. We will score the model on that result, not on the existence of AI in the drafting process.