The New Mexico Supreme Court held criminal defense attorney Stephen Aarons in direct contempt on Wednesday, fined him $5,000, and referred him to a disciplinary board after he filed an appellate brief containing fabricated witness testimony generated by ChatGPT. The court barred Aarons from appearing before it pending the outcome of discipline, struck all prior briefs from the record, and ordered the public defender's office to appoint new counsel for his former client. The underlying murder appeal will now proceed in the 2026-27 term.
Aarons, who has practiced criminal defense in New Mexico for more than 40 years, was hired in March 2025 to appeal the murder conviction of Oscar Renee Sandoval, who was sentenced to life in prison in February 2025 for killing his partner Shiereen Al-Jibury. In August 2025, Aarons filed a brief that quoted four witnesses who did not exist — Officer Michelle Amarillo, Officer Sanchez, Manal Al-Jibury, and Teresa Marquez — along with fabricated testimony attributed to Danny Stanton, Linda Stanton, and Mariah Chavez about threats and the shooter's appearance.
Aarons told the court at an August 21 hearing that he fed a Rev.com transcript of the trial, the record proper, the statement of issues, and portions of discovery into ChatGPT powered by OpenAI's o3 model, which shipped earlier in 2025. He said he did not verify the output before signing and filing it, and did not tell his client. He also inaccurately described real precedents cited in the brief, though the case citations themselves were not invented.
“I assumed that it generated a bulletproof summary of proceedings”— Stephen Aarons, New Mexico criminal defense attorney
Key facts
- 01The New Mexico Supreme Court held attorney Stephen Aarons in direct contempt and fined him $5,000 payable to the State Bar Client Protection Fund.
- 02Aarons' August 2025 appellate brief cited fabricated testimony from four witnesses who never existed, including Officer Michelle Amarillo and Officer Sanchez.
- 03Aarons fed the murder trial transcript into ChatGPT running OpenAI's o3 model and filed the output without verification.
- 04The court struck all prior briefs, barred Aarons from appearing before it pending discipline, and pushed the underlying appeal to the 2026-27 term.
- 05Justice C. Shannon Bacon cited at least eight or nine code-of-conduct provisions Aarons violated.
The fine is modest, but the collateral consequences are not. Aarons is off the case, off the court, and facing a disciplinary board that can impose suspension or disbarment. The court's order concluded he "demonstrated a lack of remorse and a lack of concern for his client."
The exchange at the August hearing was unusually direct. Justice C. Shannon Bacon told Aarons there were "at least eight or nine provisions in the code of conduct" he had violated, and rejected his claim that he did not know large language models could hallucinate facts. She noted that her 13-year-old nephew and 75-year-old stepmother both knew the problem existed.
Bacon pointed out that stories of lawyers filing AI-hallucinated cases have been running for more than three years. Aarons countered that he had submitted the brief a year ago and that "a lot has come out in the last year," a defense the bench did not accept. Bacon told him there was no material distinction between citing a fake case and describing a real one inaccurately — the duty of candor applied with equal force to both.
“So counsel, do you watch the news? Do you listen to the radio? Do you read anything about what's going on in the world? Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day”— C. Shannon Bacon, New Mexico Supreme Court Justice
Chief Justice Julie Vargas cut off Aarons' attempts to frame the hearing as an AI-policy discussion, telling him she did not care about policy concerns and cared about what to do with his client, who remained in custody. Justice Michael Vigil added that the tool did not matter: "It doesn't matter whether you use a C-student lawyer or a A-student lawyer if you didn't check their work before you filed the brief."
Bacon returned repeatedly to the harm done to Sandoval, who has now lost a year on his appeal because the briefing had to be thrown out and restarted. She compared the situation to a partner signing off on a first-year associate's fabricated work — the signature carries the liability regardless of the source.
“Your client is the one suffering because of this far more than you will ever suffer”— C. Shannon Bacon, New Mexico Supreme Court Justice
In a statement provided after the ruling, Aarons said he was remorseful and hoped the disciplinary board would treat the filing as an honest mistake. He called it "a lesson learned for all professionals who rely upon this powerful but sometimes unstable technology," and suggested the court issue a standing order requiring briefs to include a certificate of compliance regarding AI use.
The o3 model Aarons used is one of OpenAI's reasoning-tuned releases and is broadly regarded as more reliable than earlier ChatGPT versions on structured tasks. That did not save the brief. Fabricated quotes are a known failure mode when a model is asked to summarize a long document and produce witness-attributed text — the summary and the invention rarely announce themselves as separate operations to the user reading the output.
The ruling lands during a stretch of court decisions across US jurisdictions treating unverified AI output in filings as a bar-discipline matter rather than a technology curiosity. For the legal-tech vendors selling AI drafting tools into firms — Harvey, Thomson Reuters, LexisNexis, Casetext, and a growing bench of startups — the market signal is that court-facing product design has to assume the lawyer will not check the citations, and build verification into the workflow rather than trust the user to catch hallucinations downstream. Consumer ChatGPT was never that product, and rulings like New Mexico's are the price of using it as one.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
