ChatGPT-using lawyer punished for citing fake testimony from made-up witnesses

The New Mexico Supreme Court has delivered a stark warning to the legal profession: relying on artificial‑intelligence tools without rigorous verification can land a lawyer in direct contempt of court. In a Wednesday order, the state’s highest court found veteran criminal‑defense attorney Stephen Aarons guilty of submitting a brief riddled with fabricated testimony and mis‑cited authority, and it referred him to the disciplinary board while barring him from appearing before the court pending further action.

How the AI‑generated brief went awry

According to the court’s opinion, Aarons admitted that he fed a computer‑generated transcript of his client’s murder trial, along with other case documents, into a version of ChatGPT powered by OpenAI’s o3 model. The AI, which was released earlier in 2025, produced “false testimony from wholly fabricated witnesses,” including fictitious officers Michelle Amarillo and Sanchez, as well as invented civilian witnesses such as Manal Al‑Jibury and Teresa Marquez. Aarons also inserted false statements from “Danny Stanton” and “Linda Stanton” about alleged threats, and fabricated descriptions of the shooter’s clothing from “Mariah Chavez” and “Teresa Marquez.”

The brief further misrepresented legal authority, citing case law inaccurately even though the cases themselves were real. The court noted that while Aarons did not cite entirely fake cases, his inaccurate descriptions carried the same weight as outright fabricated citations under the professional code of conduct.

The court’s rebuke and disciplinary fallout

Justice C. Shannon Bacon led the court’s criticism, emphasizing that at least “eight or nine provisions in the code of conduct” were violated. She underscored that lawyers must stay informed about AI hallucinations—a point she illustrated by noting that even her 13‑year‑old nephew and 75‑year‑old stepmother are aware of the problem. Bacon asked Aarons whether he follows the news or reads about emerging technology, concluding that his ignorance was a “choice” rather than an excuse.

The court’s order barred Aarons from appearing before the New Mexico Supreme Court while the disciplinary board investigates, imposed a $5,000 fine payable to the State Bar’s Client Protection Fund, and stripped the brief from the record. The public defender’s office was instructed to appoint a new attorney for Oscar Renee Sandoval, the defendant whose murder conviction is now pending further appellate review in the 2026‑27 term.

What went wrong with the AI workflow

Aarons testified that he first used Rev.com to generate an AI‑powered transcript of the trial audio, then loaded that transcript and additional documents into ChatGPT. He claimed to have “confirmed” the transcript’s accuracy, but admitted that the error occurred after the material was fed into the language model. The court’s record shows that Aarons “did not verify the factual claims and legal authority in his AI‑generated brief before signing it and filing it,” and he failed to inform his client of these deficiencies.

This sequence highlights a common pitfall: treating AI output as a “bullet‑proof summary” without independent verification. Aarons himself said he assumed the model would be accurate because of its widespread adoption in legal and medical fields, a belief the court found untenable given the known propensity of large language models to hallucinate.

Broader implications for the legal profession

The Aarons case joins a growing docket of incidents where attorneys have been caught citing fabricated cases or misrepresenting precedent. While Aarons did not invent entire cases, the court stressed that “there’s no material distinction” between inaccurately describing a real case and citing a fake one. The ethical duty of candor to the court applies with equal force, regardless of the source of the error.

Legal scholars have warned that AI tools can amplify existing pressures on lawyers to produce faster, more voluminous briefs. The New Mexico Supreme Court’s decision serves as a concrete reminder that technology cannot replace the attorney’s responsibility to verify every fact and citation. Failure to do so not only jeopardizes a client’s case but also threatens the lawyer’s license.

Reactions from the legal community

Lawyers across the state expressed a mixture of concern and resignation. Many noted that AI‑generated drafts are now routine, but they also pointed out that the technology’s “hallucination problem” is well documented. Some bar association members called for mandatory training on AI verification, while others warned that overly punitive measures could stifle innovation.

Justice Bacon’s remarks about everyday awareness of AI hallucinations—citing her nephew and stepmother—resonated with younger attorneys who have grown up with the technology. The consensus emerging from post‑hearing discussions is that the profession must develop clear standards for AI‑assisted drafting, akin to existing protocols for electronic discovery and document management.

What this means for Oscar Renee Sandoval

Sandoval, convicted of killing his partner Shiereen Al‑Jibury in February 2025 and sentenced to life imprisonment, now faces a new appellate brief prepared by a public defender. The Supreme Court ordered that all previous briefs be struck from the record, effectively resetting the procedural slate for the 2026‑27 term. While the court did not comment on the merits of Sandoval’s appeal, the removal of Aarons’ defective brief removes a significant procedural hurdle.

Sandoval’s case underscores the human cost of AI errors. A brief riddled with fabricated testimony could have misled the court and jeopardized a fair review of his conviction. The court’s decisive action ensures that the appellate process proceeds on a factual foundation, albeit after a costly detour.

Looking ahead: safeguards and accountability

The New Mexico Supreme Court’s order signals a shift toward stricter accountability for AI use in legal practice. By referring Aarons to the disciplinary board, the court emphasizes that negligence in verification is not a mere technical slip but a breach of professional ethics. The $5,000 fine, though modest relative to potential disciplinary sanctions, serves as a financial deterrent.

Bar associations nationwide are likely to watch this case closely. Expect to see new guidelines mandating double‑checking of AI‑generated content, mandatory disclosures to clients when AI tools are employed, and perhaps even certification programs for lawyers who wish to integrate AI responsibly. Until such standards become entrenched, the onus remains on each attorney to treat AI output as a draft, not a final product.

This article was produced with AI-assisted research and editorial support. Reporting is based on the source material cited below. Sources: Ars Technica; arstechnica.com; Global1.News (12 September 2026).

By Jessica Ali, Staff Writer