The New Mexico Supreme Court has fined a Santa Fe lawyer $5,000 and held him in contempt after finding that a brief in a murder appeal contained police testimony and witnesses fabricated by OpenAI's ChatGPT.
Stephen Aarons had represented Oscar Renee Sandoval in an appeal against his murder conviction and life sentence. The court said Aarons failed to check the accuracy of material submitted in the filing, which included false testimony from witnesses who did not exist.
The invented material included purported statements describing the shooter as wearing dark pants and a white shirt. According to the court, those statements were among the false police testimony placed before it.
At a hearing on August 21, Aarons told the court that he had given ChatGPT a computer-generated transcript and other case materials. He said he had assumed the tool would produce a bulletproof summary.
Aarons later said he had not understood the extent to which artificial intelligence could hallucinate facts, meaning it can produce plausible information that is false. In a statement, he said: “I am remorseful but hopeful that the disciplinary board takes into account it was an honest mistake.”
The court panel said Aarons had shown a lack of remorse and insufficient concern for his client. It will refer him to an attorney disciplinary board for investigation. The brief does not establish what action the board may take, or whether the fabricated material affected Sandoval's conviction or the prospects for his appeal.
The appeal was assigned on September 2 to Kim Chavez Cook, a New Mexico public defender.
Justice C. Shannon Bacon addressed the wider problem at the August hearing. “The problem with lawyers relying on AI hallucinations is an above-the-fold story every single day,” she said.
The sanction is the latest in a growing number of cases in which state and federal judges have disciplined lawyers who submitted court documents produced with artificial intelligence tools without properly reviewing them. The New Mexico case shows the direct legal risk when a lawyer treats machine-generated text as a reliable account of a record rather than checking the underlying material.
The court's ruling concerns Aarons's filing and his duty to verify it. The available information does not say whether ChatGPT's training or design led to these particular false details, how commonly such tools are used in legal briefs without checks, or whether New Mexico lawyers now have specific safeguards or guidance governing their use.
Comments (0)