New Mexico lawyer sanctioned for ChatGPT-generated fake witnesses in m

Lawyer sanctioned for AI-generated false testimony in court filing

Stephen Aarons, a Santa Fe, New Mexico lawyer, was fined $5,000 and held in contempt of court after filing a murder-case appellate brief containing wholly fabricated witnesses and police testimony generated with help from ChatGPT, according to the New Mexico Supreme Court on September 11, 2026 Reuters.

Aarons was representing Oscar Rene Sandoval in an appeal of Sandoval's murder conviction and life sentence Reuters. The New Mexico Supreme Court determined that the brief included made-up police testimony and witnesses that Aarons failed to verify before submission Reuters. Aarons acknowledged he had prepared the brief with assistance from the AI program Reuters.

Court expresses escalating concern over AI vetting failures

The New Mexico Supreme Court had previously expressed exasperation at attorneys who failed to vet AI research outputs, according to a Reuters follow-up report on September 17, 2026 Reuters. The Aarons case represents a documented instance of an agent—ChatGPT—producing legally consequential false material (fabricated witness names, fictional testimony, invented police statements) that was submitted to a court of record without human verification, violating fundamental evidentiary and ethical standards.

The $5,000 fine and contempt finding are the first documented sanctions in the AI agent economy for a lawyer's use of a generative AI system to produce false legal evidence. The case demonstrates that deployed AI agents are now capable of generating materials with sufficient surface plausibility to pass initial review by a licensed attorney, yet containing fabricated facts presented as real testimony. This breach falls squarely within the "agent actions in the real world" beat: fraud committed (false testimony), a victim (the appellant and the court system), and a concrete judicial response (sanction, fine, contempt finding).

The incident also underscores a critical vulnerability in professional workflows where agents are integrated without verification protocols—a pattern likely to shape future court rules, bar association guidance, and liability frameworks governing AI use in legal practice.