A New Mexico attorney has been fined $5,000 and held in contempt by the state's Supreme Court after submitting an appeal containing AI-fabricated witnesses and fictitious police testimony in a murder case, according to The Verge, which cited a report from Reuters. The attorney, Stephen Aarons, was sanctioned for failing to meet his obligations after the fabricated material was discovered in a filing related to his client's murder conviction.
The case lands in a line of incidents that has been growing longer, and more alarming, since generative AI tools became widely accessible to the public. The legal profession was one of the first to feel the specific danger these tools pose in high-stakes professional contexts, because law involves a precise relationship with documented fact. Cases, citations, precedents — they either exist or they do not. There is no meaningful middle ground, and courts have no tolerance for material that cannot be verified, because the entire system of adversarial justice depends on both sides working from the same factual foundation. When AI models generate plausible-sounding but entirely invented case citations, witness names, or testimony, they are not making minor errors of the kind any overworked professional might make. They are introducing fiction into a process that can determine whether a person spends decades in prison.
The episode that first brought wide public attention to this problem came in 2023, when a pair of New York attorneys submitted a brief to a federal court containing citations to cases that did not exist, generated by ChatGPT. That case, involving a personal injury claim, resulted in sanctions against the lawyers involved and a wave of judicial notices across the country warning attorneys about the risks of using AI tools without rigorous verification. Several federal and state courts subsequently introduced or proposed rules requiring attorneys to disclose when AI assisted in drafting filings, and to certify that any cited authorities had been independently confirmed.
What distinguishes the New Mexico case, as reported by The Verge, is the nature of the fabricated content. Inventing case citations is serious, but those errors are relatively easy to catch during routine citation checks. Fabricating witnesses and police testimony represents a different category of problem. Witnesses are people. Testimony is a record of what someone said, under oath or on the record, at a specific time. If an AI system is generating not just phantom legal citations but phantom human beings and phantom statements attributed to law enforcement, the potential for corrupting a legal proceeding is substantially greater. In a murder appeal, where the stakes for the defendant are as high as they get in civilian law, this is not an abstract concern.
The likely consequences ripple outward in several directions. For Aarons himself, a $5,000 fine and a contempt finding are serious professional marks, though the more lasting damage may be reputational. For the broader legal community, this case will almost certainly be cited in continuing legal education contexts and in the internal guidance many firms and public defender offices are scrambling to produce. Courts are watching this pattern closely, and the judicial response has been trending toward stricter sanctions rather than leniency, on the theory that professionals who use tools they do not understand, in contexts where errors cause serious harm to real people, cannot plead ignorance as full mitigation.
There is also a question about what this means for defendants. In a murder case, the appearance of fabricated evidence in an appeal filing — even fabricated evidence introduced by the defense attorney on the defendant's behalf — could complicate any future legitimate claims the defendant might have. Courts are understandably protective of their processes, but they also retain obligations to defendants who may have had their appeals undermined not by the strength of the case against them but by their own attorney's catastrophic lapse.
The likely reading of where this goes institutionally is that courts will continue to tighten requirements, and bar associations will face growing pressure to issue clearer formal guidance rather than leaving attorneys to navigate a rapidly changing technological landscape without explicit professional standards. Some jurisdictions are already ahead of others on this. The patchwork nature of the current response is itself a problem, because attorneys practicing across state lines face different obligations depending on where they file.
What to watch next is whether bar associations move from issuing cautionary guidance to imposing formal rules with teeth, and whether courts begin distinguishing between different categories of AI-related error — treating fabricated human witnesses or testimony more severely than phantom case citations. The New Mexico Supreme Court's decision to pursue contempt, not merely a fine, suggests courts are prepared to treat this as a conduct issue rather than simply an honest mistake. If that posture hardens across jurisdictions, attorneys who have been slow to take the warnings seriously may find that the professional cost of a single filing rises considerably.
