A New Mexico defense lawyer has been fined and held in contempt after a court found that his appeal brief contained fictional witnesses and police testimony that did not exist.
Stephen Aarons, who was challenging a murder conviction, acknowledged that he had used ChatGPT while preparing the filing. The New Mexico Supreme Court said the material submitted to the court included entirely fabricated testimony and ordered Aarons to pay a $5,000 fine.
The justices also referred the attorney to the state’s disciplinary authorities for further investigation.
According to the court, the brief contained invented accounts from supposed witnesses, including a fictional description of a shooter wearing dark pants and a white shirt. The court said Aarons failed to verify the information before presenting it in a legal filing.
Aarons said he had relied on ChatGPT while reviewing the case after agreeing to handle the appeal last year. He said he supplied the AI system with a computer-generated transcript and other case documents, expecting it to produce what he described as a reliable summary of the proceedings.
Instead, the system generated details that were not supported by the record.
Aarons told the court that he had not fully appreciated how easily AI systems could produce false information. He described the incident as an honest mistake and expressed regret over what happened.
The case involves Oscar Renee Sandoval, who was convicted of murdering the mother of his children and sentenced to life in prison. His appeal remains pending and has since been assigned to New Mexico public defender Kim Chavez Cook.
The Supreme Court was particularly critical of Aarons’ failure to recognize the risks associated with AI-generated material. During an August hearing, Justice C. Shannon Bacon questioned how an attorney could remain unaware of the growing warnings surrounding AI systems and their tendency to invent information.
The case highlights a problem increasingly confronting courts across the United States. Lawyers have faced sanctions after relying on AI-generated legal research that included nonexistent cases, inaccurate quotations or incorrect statements of law.
But the New Mexico case stands out because the AI-generated material went beyond faulty legal citations. It introduced fictional witnesses and made-up testimony into a criminal appeal involving a life sentence.
The court said Aarons showed insufficient concern for the consequences of submitting the material and for the interests of his client.
Aarons, meanwhile, said the episode had taught him a difficult lesson about using AI in professional work, arguing that the technology can be powerful while still producing unreliable results when its output is not independently checked.
The appeal itself remains unresolved, while Aarons now faces a separate disciplinary review over the AI-assisted filing.
