The Supreme Court of New Mexico has held a lawyer in direct contempt after an AI-assisted appellate brief included fabricated witnesses, false testimony and misrepresented legal authorities.
The court’s order records that Stephen Aarons used ChatGPT to prepare a brief in a murder appeal but did not verify its factual claims or authorities before signing and filing it. The false material included testimony attributed to four wholly invented witnesses, as well as inaccurate evidence concerning alleged threats and the shooter’s appearance.
The consequences went beyond a routine costs order. The court struck all briefing, appointed the public defender to represent the client, barred Aarons from appearing before it pending disciplinary proceedings and directed him to pay US$5,000 to the State Bar’s Client Protection Fund.
The case is significant because the failure involved invented evidence—not merely the false case citations commonly associated with legal AI errors. Generative systems can turn incomplete records into coherent but fictitious narratives, making unverified summarisation particularly dangerous in factual and evidentiary work.
For Australian practitioners, the principle is familiar: professional responsibility remains with the lawyer who adopts and files the material. Effective AI governance must therefore verify factual propositions against the record, preserve their source locations and prevent generated assertions from being treated as evidence without human confirmation.
Sources: New Mexico Supreme Court order; Reuters.
Leave a Reply