The New Mexico Supreme Court has levied a $5,000 fine against attorney Stephen Aarons for submitting an appeal brief riddled with AI-generated falsehoods. The court held Aarons in contempt after discovering that his filing for a client’s murder conviction included testimony from witnesses who never existed and fabricated details about the shooter’s appearance. This isn’t just a clerical error; it is a direct violation of the duty to verify factual claims. The judge’s order makes it clear that relying on generative tools without human oversight is no longer a defensible excuse in legal practice.
During an August hearing, Justice C. Shannon Bacon dismantled Aarons’ defense that he simply wanted a “bulletproof summary” of the trial records. She questioned his awareness of the widespread warnings surrounding large language models. “Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what’s going on in the world?” Bacon asked, noting that AI hallucinations have been headline news daily. Her point was sharp: ignorance of a technology’s flaws is not a shield when those flaws undermine the integrity of the judicial process. Aarons admitted to using ChatGPT but claimed he did not understand the risks.
This case joins a growing list of professional stumbles involving unchecked AI use in law. Last year, two law firms faced sanctions for submitting briefs with fake legal citations, and attorneys for MyPillow’s Mike Lindell were fined for similar infractions. The pattern is becoming undeniable. As more practitioners integrate these tools to save time, courts are responding with stricter penalties. Aarons expressed remorse in a statement to Reuters, calling it an “honest mistake” and hoping the disciplinary board shows leniency. But the legal system rarely tolerates honesty when it comes at the cost of truth.
The real issue here isn’t the use of AI itself, but the abdication of professional judgment. Lawyers are paid to be skeptics, to verify every claim before it reaches a judge. When an attorney treats a probabilistic text generator as a factual database, they break that contract. This $5,000 penalty serves as a concrete warning shot. It signals that courts will not slow down their dockets to accommodate lawyers who fail to read what they file. The next step likely involves mandatory certification of human review for all AI-assisted submissions.
