A New Mexico defense attorney has been fined US$5,000 and held in contempt by the state's Supreme Court after submitting a legal brief containing fabricated witness testimony and other inaccurate information that was generated with the assistance of ChatGPT.
The case has drawn attention to the growing risks associated with the use of artificial intelligence in legal proceedings, particularly when AI generated material is submitted to a court without independent verification.
The attorney, Stephen Aarons, was representing Oscar Renee Sandoval in an appeal involving Sandoval's murder conviction. According to Reuters, Sandoval was convicted and sentenced to life in prison after being found guilty of murdering the mother of his children. Aarons later took responsibility for using ChatGPT while preparing material for the appeal.
The New Mexico Supreme Court said the appellate filing contained false testimony attributed to completely fabricated witnesses. The court also identified fictional information concerning the appearance and clothing of the alleged shooter. The material reportedly included statements suggesting that the shooter was wearing dark pants and a white shirt, even though such testimony did not exist in the actual case record.
Aarons told the court that he had provided a computer generated transcript of the trial and other case materials to ChatGPT. He said he expected the artificial intelligence system to produce what he described as a reliable summary of the proceedings.
However, the resulting material contained information that was not supported by the actual court record. Aarons admitted that he did not adequately verify the factual claims and legal authority contained in the AI assisted brief before submitting it to the court. The court subsequently criticized the failure to check the accuracy of the filing.
The Supreme Court imposed a US$5,000 fine and held Aarons in contempt. The court also decided to refer the matter to the state's attorney disciplinary board for further investigation. The disciplinary process could result in additional action, although the outcome of that process has not yet been determined.
Aarons expressed regret over the incident. In a statement reported by Reuters, he said he was remorseful and hoped the disciplinary board would consider the incident an honest mistake. He also described the episode as a lesson for professionals who use AI systems in their work.
The murder appeal itself remains pending. The case was reassigned on September 2 to New Mexico public defender Kim Chavez Cook, who is now handling the appeal. The local district attorney's office and the new public defender declined to comment on the matter, according to Reuters.
The incident highlights a broader problem that courts in the United States have increasingly encountered as lawyers adopt generative AI tools for legal research and document preparation.
AI systems such as ChatGPT can generate fluent and convincing text, but they can also produce information that appears plausible while being factually incorrect. In AI terminology, this is commonly referred to as a hallucination. In legal proceedings, such errors can have particularly serious consequences because court filings are expected to contain accurate facts, legal authorities and representations of the evidence.
The New Mexico case is more serious than many earlier examples of AI related legal errors because the disputed material reportedly included fabricated witness testimony rather than only incorrect case citations.
Courts across the United States have previously dealt with lawyers who submitted AI generated legal citations that did not exist or misquoted judicial decisions. In several cases, attorneys have faced sanctions after failing to verify material produced by AI systems. The Aarons case adds another dimension by involving invented factual testimony in a criminal appeal.
The case also raises questions about professional responsibility. Lawyers may use technology to assist with research, summarization and drafting, but the responsibility for the accuracy of material submitted to a court remains with the attorney.
AI can process large amounts of information quickly, which makes it attractive for professionals working with lengthy transcripts and legal records. However, the technology does not guarantee that every statement in a generated summary is supported by the original evidence.
The incident therefore serves as a warning for legal professionals using generative AI. Court documents should be checked against original transcripts, evidence and authoritative legal sources before they are submitted.
The case also illustrates why human oversight remains important when AI systems are used in high stakes areas. An AI generated document may appear professionally written while containing invented facts that are difficult to identify without comparing it with the underlying record.
The New Mexico Supreme Court's action against Aarons demonstrates that courts are continuing to hold lawyers accountable for the material they submit, regardless of whether the inaccurate information originated from an AI system.
For the legal industry, the case could contribute to the continuing debate over appropriate standards for using generative AI in court proceedings. Lawyers may increasingly need formal verification procedures when using AI for research, summaries or drafting.
The incident does not establish that ChatGPT independently fabricated evidence in the legal case or that OpenAI was responsible for the lawyer's court filing. Rather, according to the court and reporting on the case, Aarons used ChatGPT while preparing his filing and failed to verify the AI generated information before submitting it.
The case ultimately highlights a central limitation of generative AI in professional settings. AI can assist with drafting and information processing, but its output cannot automatically be treated as an authoritative record of evidence or law.
As courts continue to encounter AI generated errors, legal professionals are likely to face increasing pressure to establish clear procedures for checking AI assisted work before it reaches a judge.