New Mexico's Supreme Court fined a lawyer $5,000 and held him in contempt this week for filing a murder appeal that contained AI-fabricated witnesses and fake police testimony. The court found that Stephen Aarons failed to verify factual claims and legal authority in his AI-generated brief. The case adds to a growing list of legal professionals facing consequences for trusting language models without checking the output.
The brief included false testimony from witnesses who did not exist. It also contained fabricated details about the shooter's clothing and appearance. None of this material came from the actual trial record. Aarons used ChatGPT to generate what he described as a "bulletproof summary" of the case. The summary was full of holes.
A judge asked if he reads the news
During an August hearing Justice C. Shannon Bacon questioned how Aarons could be unaware of AI risks. Her questions were blunt. "Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what's going on in the world?" she asked. "Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day."
The exchange highlights a basic professional failure. Legal filing requires verification of every citation every quote and every factual claim. Aarons skipped that step entirely. He treated a language model output as finished work rather than as a starting point that needed checking.
In a statement to Reuters Aarons said he was remorseful but hopeful the disciplinary board would consider it an honest mistake. The court did not treat it that way. Contempt findings carry professional consequences beyond the fine. They go on record and can affect bar standing.
The pattern keeps repeating
Aarons is not alone. Last year a judge sanctioned two law firms for submitting a brief with numerous false inaccurate and misleading legal citations and quotations generated by AI. Lawyers representing MyPillow founder Mike Lindell were fined for putting AI-generated misquotes and fake citations in a court filing.
The problem follows a consistent arc. A lawyer uses a language model to save time on research or drafting. The model produces confident output that looks correct. The lawyer files it without verification. Opposing counsel or the court finds the fabrications. Sanctions follow.
Language models hallucinate with authority. They generate citations to cases that do not exist. They quote judges saying things they never said. They invent facts about defendants witnesses and evidence. The output reads like competent legal work until someone checks it. Most courts assume competent legal work has been checked.
Why lawyers keep making this mistake
Legal research is tedious and time-consuming. A single appeal might require reviewing hundreds of pages of trial transcripts case law and procedural rules. A language model can produce what looks like a complete brief in minutes. The speed is seductive and the labor savings are real.
The failure is not in using the tool. It is in treating the tool as a replacement for professional judgment. A lawyer who uses AI to draft a brief still has an obligation to read every word verify every citation and confirm every factual claim. That obligation does not disappear because the text came from a model instead of a human researcher.
Aarons's case is especially concerning because it involved a murder conviction. The stakes could not be higher. An innocent person could remain in prison based on fabricated evidence. A guilty person could go free based on testimony that never happened. The court's response reflects that severity.
Courts are tightening rules
Multiple jurisdictions have updated their rules to address AI-generated filings. Some courts now require lawyers to certify that they have verified AI-assisted content. Others require disclosure of AI use in briefs. The Federal Rules of Civil Procedure already require lawyers to ensure that filings are not presented for improper purposes and that factual content has evidentiary support.
The New Mexico fine demonstrates that existing rules already cover this behavior. A lawyer has a duty of candor to the tribunal. A lawyer has a duty to verify factual claims. A lawyer has a duty to ensure legal citations are accurate. Using AI does not change these duties. It just makes it easier to violate them at scale.
For developers building legal AI tools the lesson is clear. Output needs clear disclaimers. Verification workflows need to be built into the product. A legal AI tool that produces a brief should also produce a verification checklist. The tool should flag every factual claim every citation and every quote that requires human review.
The legal profession will not abandon AI tools. The productivity gains are too significant. But the Aarons case and the cases before it establish that the profession expects verification. A language model is a drafting assistant not a replacement for professional competence. The court's message is simple. Check your work or face the consequences.