1. On a Wednesday in early September, a lawyer stood before the New Mexico Supreme Court not as an advocate arguing a brilliant legal theory, but as a man explaining how he had placed his trust in the wrong place. Stephen Aarons, a Santa Fe attorney handling a murder appeal, had submitted a brief to the court that he believed was a faithful summary of trial proceedings. Instead, the document was built in part on inventions—testimony from witnesses who never existed, descriptions of events that never happened, and details that had been dreamed up by OpenAI’s ChatGPT. The state’s highest court was not amused. It fined Aarons $5,000, held him in contempt, and referred him to the disciplinary board. But the most devastating words were not the fine. They were the justices’ conclusion that Aarons had shown “a lack of remorse and a lack of concern for his client.” It was a striking rebuke, and one that underscored a growing anxiety in courtrooms across the United States: artificial intelligence can make mistakes so convincing that even lawyers cannot always tell the difference between truth and hallucination. The ruling was not merely a personal embarrassment for one attorney. It was a warning to the entire legal profession that technology, no matter how powerful, cannot replace the fundamental duty of a lawyer to verify every word before signing it in the name of a client whose freedom may hang in the balance.
2. The story began with a conviction that carried the weight of a life sentence. Oscar Renee Sandoval had pleaded not guilty to the murder of the mother of his children, but a jury found him guilty, and a judge sentenced him to life in prison. Sandoval’s case was hardly unusual in the stacks of criminal appeals that flow through state courts—one more defendant, one more family destroyed, one more attempt to find something in the record that might change the outcome. Aarons, a private attorney who took on the appeal, was not the original trial lawyer. He came to the case after the conviction, tasked with finding legal errors that could give Sandoval another chance. It was the kind of work that demands patience, precision, and a deep respect for the record. But Aarons had a helper: ChatGPT. He said he used the artificial intelligence program to summarize the trial proceedings, feeding it a computer-generated transcript and other case materials. He assumed the AI would produce what he called “a bulletproof summary.” Instead, the chatbot produced something far more dangerous: a narrative that included fictional statements, including a claim that the shooter had been wearing dark pants and a white shirt, and, even more troubling, testimony from witnesses the court later determined were wholly fabricated. When the filing reached the justices, they noticed. They ordered Aarons to explain how such material had made its way into his primary brief in a murder appeal. The question was not whether the AI had made a mistake. Every technology fails sometimes. The question was whether Aarons had done his job as a lawyer—reading, checking, investigating—before putting his name on the document and asking the court to rely on it.
3. At the August 21 hearing, the justices made their frustration clear. One of them, Justice C. Shannon Bacon, asked Aarons a question that seemed to capture the mood of the entire panel: “Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what’s going on in the world?” The question was not a rhetorical flourish. It was a devastating reminder that for months, stories about lawyers relying on AI hallucinations had been appearing in newspapers and legal publications with unsettling frequency. Judges across the country had already sanctioned attorneys for filing briefs containing fake case citations and invented legal reasoning. Aarons, according to the court, had gone further—he had introduced fabricated witness testimony into a criminal appeal, where the stakes were measured not in dollars but in human liberty. The justices said he had “demonstrated a lack of remorse and a lack of concern for his client.” Those words carried a heavy punch. In a profession where a lawyer’s credibility is his most valuable asset, being told that you appear indifferent to your client’s fate is a profound professional wound. The court did not stop at criticism. It fined Aarons, held him in contempt, and referred him to an attorney disciplinary board for investigation. The disciplinary process could lead to suspension or even disbarment, depending on what investigators find. For Aarons, the hearing was likely a nightmare—a moment when all the quiet trust he had placed in a machine came crashing down in front of an audience of judges who had seen too many cases to accept an honest mistake as a sufficient excuse.
4. In his response to the press, Aarons did not try to hide what had happened. He acknowledged using ChatGPT to summarize the trial proceedings when he agreed to take on the appeal. He admitted that he did not fully understand the degree to which AI could “hallucinate” facts—creating information that sounds plausible but has no basis in reality. He expressed remorse, calling the episode an “honest mistake.” He said he hoped the disciplinary board would take that into account. And he offered a warning to others: “It is a lesson learned for all professionals who rely upon this powerful but sometimes unstable technology.” There is something genuinely human in that confession. Many people who use AI for their work assume it is a reliable assistant, not a source of invention. They trust the polished prose and confident tone, never suspecting that the software is stitching together words based on probability rather than truth. But the law does not forgive mistakes that harm a client’s cause. In an appeal, every word matters. A false statement can undermine the entire argument, making the appellate court question everything else the lawyer says. Aarons’s statement suggests he understood that his career had been permanently altered. He was not just a lawyer who lost a case. He was a lawyer who had broken the most basic promise of the profession: that when he stands before a court, he swears to tell the truth, and that he has used all reasonable efforts to ensure that the truth is what he presents.
5. The Aarons case is part of a much larger pattern. Over the past year, state and federal judges have disciplined dozens of lawyers for submitting court documents generated by AI tools without properly vetting them. In some cases, attorneys used ChatGPT to help write briefs and ended up citing cases that did not exist. In other cases, the software misquoted actual decisions, twisting legal doctrines into shapes that had no basis in law. Most of those errors were caught and corrected, but not before judges had to spend precious time untangling the mess. The New Mexico case stands out because it involved not just fake citations but fabricated witness testimony in a criminal appeal. That is a different category of error. It goes to the heart of the judicial system’s ability to determine what actually happened. If a lawyer cannot tell the difference between real testimony and AI-generated fiction, then the entire process of legal advocacy is threatened. OpenAI, the company that creates ChatGPT, did not immediately respond to requests for comment. But the company has repeatedly warned users that AI can produce incorrect or nonsensical information. The warnings are printed on startup screens, buried in terms of service, repeated in blog posts. Yet they are easy to ignore when the technology works so well most of the time. The question now is not whether AI can be useful to lawyers—it clearly can, in tasks like organizing documents, identifying patterns, and drafting first drafts. The question is whether lawyers can be trusted to use it responsibly when the consequences of error are so severe.
6. For Oscar Renee Sandoval, the story is far from over. His appeal is still pending, and the court has assigned it to Kim Chavez Cook, a New Mexico public defender, who declined to comment. The district attorney’s office for Dona Ana County also declined to comment, likely because the case remains in litigation. But the damage has already been done. Aarons’s filing, meant to give Sandoval a chance at freedom, has become a cautionary tale of its own, casting a shadow over the appeal and forcing a new attorney to rebuild the record from scratch. The court’s decision to sanction Aarons was not just about punishing one lawyer. It was about protecting the integrity of the entire system. Courts cannot function if they cannot trust the documents filed by the lawyers who appear before them. Judges rely on advocates to present the facts accurately, to cite the law honestly, and to flag uncertainties rather than conceal them. When a lawyer outsources that responsibility to an AI program and fails to check the result, the entire foundation of justice wobbles. The lesson of the case is not that technology is bad. It is that human judgment remains irreplaceable, especially in the high-stakes arena of criminal law, where a fabricated sentence can send a person to prison for life or overturn a conviction for the wrong reason. Aarons said he learned his lesson. Perhaps the broader legal profession is learning it too—one hallucination at a time. But for Sandoval, whose life is still at stake, the hope is that the truth, however delayed, will eventually find its way into the record, and that the court will give him the fair appeal that he deserves. In the end, the most human thing about this story is not the technology that failed. It is the reminder that justice depends on fallible people, and that no machine can ever replace the simple, irreplaceable act of a lawyer reading the words before signing them.