Legal work still carries consequences if AI tools fail verification

A CourtListener RECAP docket entry posted in late October in In Re: OpenAI, Inc. Copyright Infringement Litigation reproduces a filing that says defendants’ AI tools cannot stand in for the press because they often fail to verify the accuracy of their outputs and produce hallucinatory misinformation. The filing places verification at the center of whether AI-generated work can substitute for work that carries professional or legal responsibility. It does not impose a sanction or decide the underlying dispute, but it makes the failure to check an output the relevant fact for judging the work.

That matters for legal work delivered with AI because speed or production capacity does not remove the obligation to deliver accurate work. The CourtListener RECAP entry in Clark v. Social Security, Commissioner of shows the consequence in a legal filing: the court said generative AI was used in preparing a brief and ordered Rich to show cause why she should not be sanctioned, including with a fine, for factitious citations. The CourtListener: AI and privilege (opinions) entry in Cervantes v. Bianco likewise says the court disapproved of improper use of artificial intelligence tools that apparently contributed to the problem before it. Together, the matters move the question away from whether AI was used and toward whether the resulting legal work was reliable.

The stronger reading remains narrower than a general verdict on AI in law. The OpenAI filing says AI tools fail verification and produce misinformation; it does not establish that every AI tool does so, or that every AI-assisted legal filing is defective. The Clark order establishes that one court connected AI use in one brief with factitious citations serious enough to prompt a show-cause order. It does not establish that the use of AI itself caused every citation problem, that Rich will be fined, or that an AI-native legal model cannot produce sound work. The Cervantes opinion establishes a separate judicial disapproval of improper AI use and says that use apparently contributed to the matter before the court. It does not identify a general rule governing all jurisdictions or all forms of AI assistance. For the stronger claim to hold, courts would need to impose consequences based on demonstrated defects in AI-assisted legal work across more than one matter, while the work’s provider remained accountable for checking it. These items establish the accountability side of that proposition, not its universal scope.

The next concrete fact to watch is the order resolving the show-cause proceeding in Clark v. Social Security, Commissioner of. A sanctions order, with or without a fine, would show how the court treats factitious citations after AI-assisted preparation was disclosed in the matter. A decision imposing consequences for the defective work would reinforce the distinction between using AI and verifying what it produces. A decision declining sanctions would narrow what this matter shows, without erasing the filing’s warning that unverified outputs can fail at the point where legal work must be trusted. We will score the direction by the order imposed on the work, not by the presence of AI alone.