The New Mexico Supreme Court held a veteran criminal defense attorney in direct contempt this week after he submitted a brief containing fabricated witness testimony and inaccurate case descriptions generated by ChatGPT. The lawyer, Stephen Aarons, admitted he did not verify the AI-generated content before filing it with the court, and did not inform his client or the client's family about the errors.

How the fabrication happened

Aarons, who has practiced criminal defense law in New Mexico for over 40 years, was hired by family members of Oscar Renee Sandoval to appeal a murder conviction. Sandoval was sentenced to life in prison in February 2025 after being convicted of killing his partner, Shiereen Al-Jibury. In August 2025, Aarons submitted a brief that contained testimony from witnesses who never testified at trial, fake police quotes, and inaccurate descriptions of legal precedents.

At a hearing on August 21, Aarons told the court he fed a computer-generated transcript of the murder trial, generated by Rev.com, into ChatGPT along with other case materials. He said he used a version of ChatGPT powered by OpenAI's o3 model and assumed it would produce an accurate summary. The model generated fabricated testimony from witnesses including Officer Michelle Amarillo, Officer Sanchez, Manal Al-Jibury, Teresa Marquez, Danny Stanton, Linda Stanton, and Mariah Chavez, none of whom had testified at trial or were accurately represented in the brief.

"It's of little comfort to know that my stupidity is what brings us together this afternoon," Aarons told the court.

The court's response

The justices were unsparing. Justice C. Shannon Bacon told Aarons he violated at least eight or nine provisions of the code of conduct and expressed disbelief at his claimed ignorance of AI hallucinations. She noted that her 13-year-old nephew and 75-year-old stepmother both know that AI tools can generate false information.

"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?" Bacon said. "Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day. So either you buried your head in the sand, and that's a choice to do that, an intentional choice to be uninformed, or you took a gamble, and neither of those are consistent with the code of conduct."

Aarons responded that he submitted the brief a year ago and "a lot has come out in the last year." The problem of lawyers citing AI-generated fake cases in briefs has been reported regularly for more than three years, and the first high-profile incident occurred in 2023.

The court fined Aarons $5,000, to be paid to the State Bar of New Mexico Client Protection Fund, barred him from appearing before the New Mexico Supreme Court pending disciplinary proceedings, and referred him to the disciplinary board for further action. All previous briefs in the case were struck from the record, and the court ordered the public defender office to appoint a new attorney for Sandoval.

The client never learned what happened

Bacon pressed Aarons on whether he had told his client about the ChatGPT use. Aarons said he had not, explaining that Sandoval is a Spanish speaker and he "wanted to see what happened today" before disclosing the problem. He told the court he informed Sandoval's brother and cousin only that "there was a problem with the first brief," without mentioning ChatGPT or the fabricated testimony.

"You didn't say, 'I relied on ChatGPT and it submitted lies to the court and the court is mad at me and now there's a hearing,'" Bacon said. "I'm guessing you didn't tell your client's family members that."

Aarons confirmed he did not discuss ChatGPT with them. Bacon told him that Sandoval needed to know the full extent of the problem so he could make an informed decision about whether Aarons was the right person to represent him.

The verification problem

The justices rejected Aarons's suggestion that the court should issue standing orders requiring certificates of compliance regarding AI use. Chief Justice Julie Vargas told him the court was not interested in his policy proposals and wanted him to address the immediate harm to his client, who remained in custody while the appeal was delayed.

Justice Michael Vigil pointed out that the tool doesn't matter. Whether the work came from a first-year associate, a law student, or an AI model, the lawyer signing the brief bears responsibility for its accuracy. The issue wasn't that Aarons used ChatGPT. It was that he signed a brief without reading it carefully enough to notice fabricated witnesses and misquoted cases.

"It doesn't matter what the tool is. It doesn't matter how advanced the AI-generated program is or what improvements they make," Vigil said. "It doesn't matter whether you use a C-student lawyer or an A-student lawyer. If you didn't check their work before you filed the brief, that's the issue."

In his statement to Ars Technica, Aarons called it "an honest mistake" and expressed hope that the disciplinary board would consider that he had no prior awareness that AI could hallucinate facts. He said he was remorseful and hoped his case would prevent other professionals from making the same error.

The case highlights a gap between the technical capability of AI tools and the professional obligations of people who use them. AI models can generate plausible-sounding text that is factually wrong. When that text is submitted to a court as part of a criminal appeal, the consequences fall on the defendant, not the lawyer or the model. The New Mexico Supreme Court's order makes clear that the profession's existing rules already require verification, and that claiming ignorance of how the tool works is not a defense.