A New Mexico lawyer is in serious legal trouble after using ChatGPT to generate an appeal brief that contained entirely fabricated witnesses and testimony. Stephen D. Aarons, who was representing a man convicted of killing his wife, told a panel of five judges that he assumed the AI would produce a “bulletproof summary of proceedings.” Instead, the chatbot invented people, including a nonexistent Danny Stanton and Linda Stanton, who supposedly received threats, as well as other false statements about the shooter's appearance.
During an August hearing, the judges did not hide their frustration. One noted that it would be nearly impossible to avoid hearing about lawyers relying on AI hallucinations, calling the pattern an “above-the-fold story every single day.” Another judge pressed Aarons on whether he had buried his head in the sand or taken a gamble, adding that both choices are inconsistent with the code of conduct. The court ultimately found Aarons in direct contempt, fined him $5,000, barred him from appearing before the court pending an investigation, and removed him from the case.
Aarons told Reuters he was remorseful and hoped the disciplinary board would see it as an honest mistake. He also admitted he had not directly informed his client, who is serving life in prison, instead telling family members only that there was “a problem with the brief.” The case adds to a growing list of attorneys caught submitting AI-generated errors in court filings, a problem judges have called “striking, concerning, disappointing, and saddening.”