Lawyer Held in Contempt for Filing AI Brief with Fabricated Witnesses

The New Mexico Supreme Court held criminal defense attorney Stephen Aarons in direct contempt on Wednesday after he filed a ChatGPT-generated appellate brief containing testimony from wholly fabricated witnesses, including invented police testimony. Aarons, who has practiced in New Mexico for over 40 years, admitted he never verified the brief's factual claims or legal citations before filing it. He also did not tell his client. The court referred him to a disciplinary board and found he showed a lack of remorse and a lack of concern for his client.

What happened

Detail	Fact
Court	New Mexico Supreme Court
Attorney	Stephen Aarons
Finding	Direct contempt of court
Violation	Brief contained false testimony from wholly fabricated witnesses, including fake police testimony
Admission	Did not verify factual claims or legal authority in the AI-generated brief before signing and filing
Client	Oscar Renee Sandoval, sentenced to life in prison in February 2025
Next step	Referred to state disciplinary board for further proceedings

Stephen Aarons has been a criminal defense lawyer in New Mexico for over 40 years. He was hired by family members of Oscar Renee Sandoval to appeal a murder conviction. Sandoval was sentenced to life in prison in February 2025 for the killing of Shiereen Al-Jibury, his partner and the mother of his children.

Instead of conducting standard legal research, Aarons used ChatGPT to help prepare the appellate brief. The New Mexico Supreme Court found the brief included “false testimony from wholly fabricated witnesses,” among other factual and legal misrepresentations. According to the court’s order, Aarons admitted he signed and filed the document without checking whether any of it was accurate.

The court also found that Aarons never disclosed any of this to his client, meaning Sandoval had no idea the brief filed on his behalf in a life-sentence appeal was built on invented facts.

Why it matters

This is not the first time a lawyer has faced punishment for relying on AI-generated legal research without verification. Courts across the US have been dealing with the consequences of attorneys submitting briefs that cite nonexistent cases or fabricated sources, a problem commonly called AI hallucination. In this instance the stakes were higher: a man serving life in prison had his appeal undermined by a brief his own lawyer never bothered to read critically.

The New Mexico Supreme Court’s finding of direct contempt, combined with a referral to the disciplinary board, signals that courts are moving past warnings and toward concrete professional consequences. The court’s language about a “lack of remorse” suggests Aarons may face serious disciplinary action, potentially including suspension or disbarment.

For anyone using AI tools in high-stakes professional work, the lesson is the same one courts keep repeating: the professional who signs the document is responsible for what is in it, regardless of how it was produced. We covered a related case where a lawyer was fined $5,000 for AI-hallucinated citations in another murder appeal, and the pattern is becoming hard to ignore.

Our take

The tool is not the problem here. ChatGPT did what language models do: it produced confident-sounding text that was not grounded in reality. The problem is a lawyer who treated an AI output as a finished work product and filed it in the highest court in his state on behalf of a client facing life in prison.

Forty years of legal experience should have included the judgment to verify any document before signing it, AI-generated or otherwise. The fact that he also withheld this from his client makes it worse. Courts are clearly done treating AI hallucination as a novelty or an honest mistake when basic verification steps were skipped entirely.

If your business is exploring AI integration for any client-facing or compliance-sensitive work, build a verification step into the workflow before anything goes out the door. AI drafts; humans sign off. That separation has to be real, not nominal.

What to do about it

1. Treat every AI-generated output as a draft, not a finished document, especially in legal, financial, or compliance contexts.
2. Assign a specific person to verify facts, citations, and sources before any AI-assisted document is signed or submitted.
3. Log what tools were used to produce each document so there is an auditable trail if questions arise later.
4. Disclose AI use to clients or stakeholders where relevant, particularly when the stakes involve legal proceedings or binding decisions.

If you are building AI into your business workflows and want to get the governance side right from the start, talk to the Lumien team before the process is already in production.

Source: Ars Technica · AI