Stephen Aarons, who has practiced criminal defense law in New Mexico for more than 40 years, was sanctioned for using fabricated testimony and inaccurate legal citations generated by ChatGPT in an appellate brief concerning a murder conviction. The New Mexico Supreme Court found that Aarons had failed to verify the factual and legal claims in the brief before signing and filing it, and had also failed to inform his client of this.
Aarons was handling Oscar Renee Sandoval’s appeal. Sandoval had been convicted of killing Shiereen Al-Jibury, his partner and the mother of his children, and was sentenced to life in prison in February 2025. In a brief filed in August 2025, Aarons used statements as if they belonged to witnesses who had never been called at trial. The court said the brief included fabricated testimony attributed to “entirely fictitious” witnesses named Officer Michelle Amarillo, Officer Sanchez, Manal Al-Jibury and Teresa Marquez, as well as inaccurate statements attributed to Danny Stanton, Linda Stanton and Mariah Chavez. Legal rulings concerning real cases were also misrepresented in the brief.
At a hearing on August 21, Aarons said he had uploaded the computer-generated trial transcript in the murder case and other documents in the file to ChatGPT. Aarons, who said he had used a version of ChatGPT powered by OpenAI’s o3 model, released in early 2025, stated that he did not know artificial intelligence could “hallucinate facts” and had assumed the system had prepared a reliable summary. However, he acknowledged that he had written the brief himself and that it contained numerous errors in the table of contents and summary section.
The court struck all of Aarons’ previous briefs from the file and ordered that a new attorney be appointed for Sandoval. The case will be heard during the court’s 2026-27 term. Aarons was barred from practicing before the Supreme Court until the disciplinary investigation is concluded; he was also fined $5,000, payable to the New Mexico State Bar Client Protection Fund. Additional sanctions may be imposed during the disciplinary process.
The justices emphasized that the use of artificial intelligence is not in itself a problem, but that the attorney who signs the brief must verify the accuracy of its contents. Aarons acknowledged that he did not tell Sandoval about the problem and told his family only that there was an issue with the brief. Chief Justice Julie Vargas, meanwhile, said the situation had serious consequences for the incarcerated client.
Why it matters
The ruling shows that, rather than banning the use of generative artificial intelligence in law, responsibility for content prepared with these tools remains with the attorney. This distinction means that every piece of information entered into the file must be independently verified, particularly in proceedings such as appeals that seek to have a conviction reconsidered. When no verification is carried out, the resulting risk is not limited to professional reputational damage or disciplinary sanctions; it can also affect the credibility of an incarcerated person’s filing and their right to legal assistance. How Sandoval’s case will proceed with the new attorney, and whether additional sanctions will be imposed in the disciplinary process against Aarons, remain two key unresolved issues.
Background
ChatGPT is not a new name in the FikirPilot archive: over the past 90 days, we have published 17 articles mentioning the name; the latest is dated September 10, 2026.