New Mexico Lawyer Cited ChatGPT-Fabricated Witnesses in Murder Appeal

New Mexico attorney Stephen D. Aarons used ChatGPT to generate briefs while representing a man accused of shooting his wife. [1] According to a New Mexico Supreme Court direct-contempt order, the briefs contained false testimony from “wholly fabricated witnesses.” [1]

The fabricated material included a nonexistent Danny Stanton saying he received threats, a made-up Linda Stanton saying her husband received threats, and statements from other AI-generated people about the shooter’s clothing and appearance. [1] Aarons told five judges that he had assumed ChatGPT would produce a “bulletproof summary of proceedings,” blamed the outcome on his own “stupidity,” and separately told Reuters that he was remorseful and considered it an honest mistake. [1]

The central failure was not only that ChatGPT produced false material, but that fabricated witnesses and testimony entered briefs after Aarons assumed the output would be reliable. Because the errors concerned named witnesses, alleged threats, and the shooter’s appearance, checking the generated material against the proceedings would have tested specific assertions before filing. Aarons’s remorse does not resolve the workflow question exposed by the contempt order: who must review AI-assisted drafting and how that review should be documented. What the sources don't address: whether the client’s appeal was affected or what discipline Aarons may ultimately face.