New Mexico's Supreme Court fined lawyer Stephen Aarons $5,000 and held him in contempt for filing an appeal in a murder case that cited witnesses who do not exist. The brief, prepared with ChatGPT, contained fabricated testimony and fake police statements, and the court found that Aarons failed to verify the factual claims and legal authority in the document. The case shows how quickly an AI-generated draft can turn into a formal sanction when no one checks it against the record.
What the court found in the brief
The filing states that the brief «contained false testimony from wholly fabricated witnesses», along with false testimony about the shooter's clothing and appearance. The court held Aarons in contempt specifically for failing to verify the factual claims and legal authority in his AI-generated brief, according to a report from Reuters. The penalty was issued in a filing on Wednesday. The fabricated material was not a minor citation error: invented people and invented police statements were presented to the court as part of an appeal against a murder conviction.
During an August hearing, Aarons admitted he had used ChatGPT and expected it to produce what he described as a «bulletproof summary» of the trial. Justice C. Shannon Bacon challenged that assumption directly. «Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what's going on in the world?» she said, according to Reuters. She added that the problem of lawyers relying on AI hallucinations is «an above-the-fold story every single day». The exchange points to a gap between a tool that produces fluent text and a professional duty to confirm that the text is true.
The mechanism behind the failure is straightforward. A language model generates citations, names and quotations that read as plausible because they follow the patterns of real legal writing, but it does not check them against a case file or a court record. Verification is the step that catches this, and in Aarons's brief that step was skipped. The court's contempt finding treats the omission as the violation, not the use of ChatGPT itself. For any team using AI to draft documents, the practical rule is that the model produces a draft and a person remains responsible for every factual claim in it.
The sanction lands in a widening pattern. The number of lawyers using AI tools in their work is increasing, and so are the instances in which a court discovers citations that are fabricated or hallucinated by AI. Last year, a judge criticized two law firms for submitting a brief with «numerous false, inaccurate, and misleading legal citations and quotations». Lawyers for MyPillow's Mike Lindell were also fined for putting AI-generated misquotes and fake AI citations in a brief. Courts are now treating these filings as a recurring category of misconduct rather than isolated mistakes.
What this means for businesses using AI
For companies that use AI to prepare contracts, regulatory filings, audit notes or sales proposals, the case sets a clear boundary: the tool can speed up drafting, but the review burden does not move to the model. A small firm without a legal department is exposed most, because the same person writes and checks the document, and a fluent paragraph is easy to accept. A larger organization can separate drafting from verification and assign a second reviewer, which turns the same AI output into a lower-risk input. The cost difference between the two setups is one review pass.
What the ruling does not establish is that AI drafting is prohibited, or that ChatGPT specifically caused the outcome. The court acted on unverified claims in a filed document, and the fine of $5,000 reflects that procedural failure. Before adopting an AI writing tool for regulated work, a buyer should ask the vendor how the system handles citations and source attribution, whether it can link each claim to a retrievable document, and what the audit trail looks like when a claim turns out to be wrong. Those questions matter more than benchmark scores for a team that signs its name to the output.
Aarons told Reuters he is «remorseful but hopeful that the disciplinary board takes into account it was an honest mistake». The marker to watch is the disciplinary board's response: if it treats the case as an isolated lapse, verification stays a matter of individual discipline, and if it treats unverified AI content as a standing risk, firms will start writing review steps into their procedures. For business buyers of AI tools, the second outcome is the one that changes how contracts and internal policies are drafted.
