A New Mexico defense attorney has been fined $5,000 and held in contempt after filing a murder appeal containing fabricated witnesses and police testimony generated by ChatGPT. Attorney Stephen Aarons used the AI system to summarize trial materials in the appeal of Oscar Renee Sandoval, who was convicted of murder and sentenced to life imprisonment, but failed to verify the resulting claims before submitting them to the New Mexico Supreme Court. The court found that the brief contained false testimony attributed to entirely fictional witnesses, including invented descriptions of the alleged shooter. Aarons said he believed ChatGPT would produce a reliable summary and did not understand the extent to which artificial intelligence could “hallucinate” information. The episode underscores a basic principle increasingly important as AI enters professional life: technological sophistication does not relieve lawyers—or anyone else entrusted with consequential decisions—of responsibility for verifying facts before acting upon them.
Key Takeaways
- The New Mexico Supreme Court fined attorney Stephen Aarons $5,000 and held him in contempt after an AI-assisted appellate brief contained fabricated witnesses and nonexistent police testimony.
- Aarons acknowledged using ChatGPT to summarize trial records without independently verifying its output, saying he did not appreciate the technology’s capacity to invent convincing but false information.
- The case goes beyond AI-generated bogus legal citations because the system effectively manufactured evidence in a murder appeal, demonstrating why human accountability and verification must remain paramount when AI is used in courts and other consequential institutions.
In-Depth
A New Mexico attorney’s reliance on ChatGPT has produced a striking warning about artificial intelligence in the legal system: sophisticated software cannot substitute for professional responsibility. Stephen Aarons, representing Oscar Renee Sandoval in an appeal from a murder conviction and life sentence, used ChatGPT to summarize trial materials. The resulting appellate brief contained testimony attributed to witnesses who did not exist, along with false statements concerning the alleged shooter’s appearance.
The New Mexico Supreme Court held Aarons in contempt and imposed a $5,000 fine after concluding that he failed to verify the AI-generated material before submitting it. The court also indicated that the matter would be referred for disciplinary review. Aarons characterized the episode as an honest mistake and said he had not appreciated how extensively an AI system could hallucinate facts.
The episode is more serious than the increasingly familiar problem of AI inventing legal citations. Fabricating testimony effectively inserts nonexistent evidence into a judicial record, and in a criminal appeal the consequences could affect a defendant facing life imprisonment. That makes verification not merely a technical precaution but an essential professional obligation.
Artificial intelligence can accelerate research, organize records and help lawyers identify arguments, but speed is worthless when accuracy disappears. The broader lesson extends beyond law. AI systems generate plausible language, not guaranteed truth. Institutions should resist treating polished output as verified fact. Technology may assist human judgment, but accountability must remain with the person who signs the brief, publishes the report or makes the consequential decision.
Sources
- https://www.reuters.com/legal/government/chatgpt-invented-fake-police-testimony-murder-appeal-new-mexico-high-court-says-2026-09-11/
- https://supremecourt.nmcourts.gov/about-this-court/recordings-of-oral-arguments/
- https://www.theverge.com/ai-artificial-intelligence/994207/chatgpt-new-mexico-lawyer-fined-murder-appeal