Court fines lawyer for citing fake OpenAI o3 testimony

The New Mexico Supreme Court has found veteran defense attorney Stephen Aarons in contempt after he submitted an appellate brief containing fictional witness statements produced by OpenAI's ChatGPT o3 model. Aarons, who has practiced law for more than 40 years, utilized the AI tool to draft an appeal for his client, Oscar Renee Sandoval. Sandoval received a life sentence in February 2025 for the murder of Shiereen Al-Jibury. Aarons admitted he failed to cross-check the facts and legal precedents in the document before filing it in August 2025.

To prepare the filing, Aarons processed trial audio through the transcription service Rev.com and fed the resulting text, along with other case discovery, into ChatGPT. The OpenAI o3 model, which debuted earlier in 2025, hallucinated extensive portions of the trial. The brief cited fake testimony from nonexistent witnesses, including Officer Michelle Amarillo, Officer Sanchez, Manal Al-Jibury, and Teresa Marquez. It also included fabricated statements attributed to real witnesses, such as Danny Stanton, Linda Stanton, and Mariah Chavez, and mischaracterized several actual legal precedents.

During an August 21 hearing, state Supreme Court justices strongly criticized Aarons for failing to review the AI-generated output. Justice C. Shannon Bacon noted that Aarons violated multiple professional conduct rules, rejecting his defense that he was unaware AI could hallucinate facts. As a result, the court fined Aarons $5,000, payable to the State Bar of New Mexico Client Protection Fund, and barred him from appearing before the state Supreme Court pending a disciplinary board investigation. The court also struck the tainted briefs and ordered a public defender to take over Sandoval's appeal for the 2026-27 term.

For legal practitioners, this ruling reinforces that the adoption of advanced AI models does not absolve professionals of their ethical duties. While tools like ChatGPT o3 can streamline document analysis, the responsibility for factual accuracy remains solely with the human signer. Relying blindly on AI summaries without rigorous verification can lead to severe sanctions, damaged client trust, and potential disbarment.