New Mexico Supreme Court Makes History With AI Contempt Ruling
On Wednesday, September 10, 2026, the New Mexico Supreme Court issued an order that reverberated through every law firm in the country: a veteran criminal defense attorney was held in direct contempt for filing an AI-generated appellate brief stuffed with "false testimony from wholly fabricated witnesses." The ruling — the first of its kind at a state supreme court level — marks a categorical escalation in how the judiciary is responding to lawyers who outsource their professional judgment to ChatGPT and similar tools without exercising any meaningful oversight.
The court did not stop at contempt. It referred attorney Stephen Aarons to the state's attorney disciplinary board for further proceedings, a move that could end his ability to practice law. In a profession where reputation is currency, the New Mexico Supreme Court's language was damning: Aarons had "demonstrated a lack of remorse and a lack of concern for his client."
This is the ruling that AI-using attorneys have feared since 2023. It is also, legal ethics scholars argue, entirely foreseeable.
The Lawyer at the Center: Stephen Aarons and the Sandoval Appeal
Stephen Aarons is not a young associate dazzled by new technology. He has practiced criminal defense law in New Mexico for more than four decades — a career spanning thousands of hearings, motions, and appellate filings. That makes the facts here particularly striking.
Aarons was retained by family members of Oscar Renee Sandoval, a man convicted of murdering Shiereen Al-Jibury, the mother of his children. Sandoval was sentenced to life in prison in February 2025. His family sought an appeal, hoping Aarons's experience could identify reversible error in a case with stakes that could not be higher.
Instead, the brief Aarons filed contained fabricated witness testimony — including fake accounts attributed to law enforcement officers who apparently do not exist — along with multiple factual and legal misrepresentations. When the court confronted him, Aarons admitted he had not verified a single factual claim or legal citation in the AI-generated document before signing it and submitting it under his professional seal. He also admitted he never told Sandoval that the brief was AI-generated or that it contained serious inaccuracies. His client, facing life imprisonment, was kept entirely in the dark.
The court's finding of direct contempt — a civil or criminal sanction imposed for conduct that obstructs or disrespects the judicial process — reflects how seriously the justices viewed that combination of negligence and silence.
Why This Ruling Is Different From Prior AI-Citation Cases
The legal profession's reckoning with AI-generated hallucinations began in earnest in June 2023, when federal Judge P. Kevin Castel sanctioned attorneys in Mata v. Avianca, Inc. (S.D.N.Y.) for citing a string of fictitious cases generated by ChatGPT. That case became the cautionary tale circulated at every bar association CLE session and law school orientation that followed. The lawyers were fined and ordered to notify the judges whose nonexistent opinions they had cited. It was humiliating. It was expensive. It was also, critics now argue, too mild to change behavior at scale.
The New Mexico ruling operates in an entirely different register. Mata v. Avianca produced Rule 11 sanctions under the Federal Rules of Civil Procedure — a mechanism designed to deter bad-faith or reckless filings. Direct contempt carries a different legal and symbolic weight. It is the court declaring that an officer of the court has actively undermined the administration of justice, not merely filed a sloppy document. The referral to a disciplinary board introduces the possibility of suspension or disbarment.
There is also a factual distinction that elevates the severity. The Mata lawyers cited fake cases — phantom precedents that didn't exist. Aarons's brief included fabricated witness testimony. Inventing a case citation is a serious research failure. Submitting what amounts to false sworn testimony — even inadvertently, through unreviewed AI output — strikes at the evidentiary foundation of the legal system itself.
The Professional and Ethical Fallout for AI-Using Attorneys
The American Bar Association's Model Rule 1.1 requires lawyers to provide competent representation, which includes the legal knowledge, skill, thoroughness, and preparation reasonably necessary for the representation. A 2012 amendment to the official commentary added that competence encompasses staying current with changes in the law and its practice, including the benefits and risks associated with relevant technology.
That commentary has become the lodestar for state bar advisories on AI use. By the close of 2025, more than a dozen state bars had issued formal guidance — ranging from California and New York to Florida and Illinois — warning that attorneys who use generative AI without verification may be in breach of their competence obligations. The ABA itself issued Formal Opinion 512 in 2024, clarifying that lawyers must understand the limitations of AI tools, must supervise their use, and must independently verify outputs before filing.
Aarons appears to have done none of this. His admission that he signed and filed the brief without verifying its contents is, on its face, an admission of a Rule 1.1 violation. The disciplinary board referral will determine whether that failure rises to the level of professional misconduct warranting formal sanction.
For the broader bar, the chilling effect is real. The question is whether it produces more careful AI use or a reflexive retreat from tools that, when properly supervised, can genuinely improve access to justice and research efficiency.
What Courts and Legal Bodies Are Doing to Govern AI Use
Courts have not waited for bar associations to move. Since 2024, dozens of federal and state courts have adopted standing orders requiring attorneys to disclose whether AI tools were used in drafting filings and to certify that the contents were independently reviewed for accuracy. Some judges, particularly in the federal district courts, now attach specific AI-disclosure requirements to every scheduling order.
The New Mexico Supreme Court's contempt ruling goes further than disclosure. It treats the failure to verify AI output as an act with direct consequences — not just a procedural oversight to be corrected on resubmission. The referral mechanism sends a clear signal: the court views AI negligence in high-stakes filings as a disciplinary matter, not merely a technical one.
State supreme courts occupy a unique position in this conversation. They govern attorney discipline within their jurisdictions and set the ethical standards that trial courts apply. A contempt finding at that level — accompanied by a disciplinary referral — creates persuasive authority that other courts can cite when facing similar misconduct. It is, in the truest sense of the phrase, a precedent.
Lessons for Lawyers: How to Use AI Tools Without Risking Contempt
The Aarons case is not an argument against using AI in legal practice. It is an argument against using AI as a replacement for professional judgment. The distinction matters enormously.
Competent AI use in a law practice looks like this: a lawyer uses a tool to draft an initial brief, then reads every sentence, verifies every citation against primary sources, checks every factual claim against the case record, and — critically — discloses to the client that AI was involved and what review process was applied. The AI output is a starting point, not a finished product.
What Aarons did was the opposite. He treated the AI's output as reliable without checking it, signed his name to it as a professional attestation, and filed it in a capital case affecting a man's freedom. That a 40-year veteran of the bar made this mistake underscores that experience alone does not confer immunity to the seductive efficiency of generative AI.
The courts are now making the stakes explicit. Direct contempt. Disciplinary referral. Public rebuke from a state's highest court. For lawyers who have not yet implemented a formal AI-verification workflow, the message from Santa Fe is unambiguous: the time to build that workflow is before the next brief is filed, not after the next contempt order lands on your desk.
Source: [Ars Technica - All content](https://arstechnica.com/tech-policy/2026/09/chatgpt-using-lawyer-punished-for-citing-fake-testimony-from-made-up-witnesses/)