New Mexico Supreme Court fines lawyer $5,000 over AI-invented witnesses in murder appeal

Court imposes sanctions

New Mexico's highest court has fined a defence attorney $5,000 and held him in contempt after an appeal he filed for a client convicted of murder was found to rely on witnesses and testimony that an AI tool had invented.

According to The Verge, which cites a Reuters report, the New Mexico Supreme Court imposed the sanction in a filing on Wednesday. The court concluded that lawyer Stephen Aarons had failed to check the factual assertions and legal authority in a brief he generated with AI before submitting it.

The brief, the court said, presented testimony from witnesses who did not exist at all, along with invented claims about the clothing and appearance of the shooter.

The August hearing

At a hearing in August, Justice C. Shannon Bacon pressed Aarons on why he was unaware of a risk that has become a staple of news coverage. As Reuters reported, she opened with "Counsel, do you watch the news?" and asked whether he listened to the radio or read about current events at all, adding that lawyers relying on AI hallucinations now make front-page news nearly every day.

Aarons acknowledged that he had used ChatGPT for the filing and said he had assumed it would produce a "bulletproof summary" of the trial.

In a statement to Reuters, Aarons said he regretted what happened and hoped the disciplinary board would take into account that it was an honest mistake. His mention of the board suggests the episode may not end with Wednesday's fine.

Courts keep catching the same failure

The New Mexico case adds to a lengthening list of AI-related courtroom errors. The Verge points to an incident last year in which a judge rebuked two law firms for a brief containing numerous false and misleading legal citations, and to lawyers for MyPillow's Mike Lindell being fined over AI-generated misquotes and fabricated citations in a filing. The Verge also notes that as more lawyers adopt AI tools, courts are uncovering such problems more often.

But there is a difference in kind here. Most earlier incidents involved hallucinated case law — decisions and quotations that look genuine but cannot be found. In Aarons' appeal, according to the court, the fabrications extended to the facts of the case itself: invented people presented as witnesses and false descriptions of the shooter's appearance, in a matter where a murder conviction was at stake.

Why it matters

The sanction shows that hallucination is not confined to citations. Language models also invent real-world facts, and when those enter a legal filing they can misrepresent the record of the case itself — a more fundamental failure than citing a non-existent opinion.

It also confirms where responsibility sits. Courts have signalled that using AI tools is acceptable; submitting their output unverified is not. The penalty here — a $5,000 fine plus a contempt finding, with further disciplinary review possibly ahead — shows consequences hardening from warnings into material punishment.

For the client, the fallout can be severe: fabricated evidence claims can sink an appeal on their own and taint whatever legitimate arguments existed alongside them.

And Justice Bacon's line of questioning makes clear that not knowing about AI hallucination is no longer an excuse. The risk is treated as common knowledge within the profession, and lawyers are expected to verify everything a model produces before a court ever sees it.