AI Newsway

New Mexico's Top Court Fines a Lawyer $5,000 Over ChatGPT's Fake Witnesses

The court found fabricated testimony, not just the usual invented citations

|5 min read0
AI Summary
New Mexico's Supreme Court fined Santa Fe attorney Stephen Aarons $5,000 and held him in contempt after finding his appellate brief quoted witnesses and police testimony that ChatGPT had invented. Aarons said he fed a trial transcript to the chatbot expecting a reliable summary. The case escalates a growing pattern of AI sanctions because the fabrications were testimony, not citations, in an appeal by a defendant serving a life sentence.
A courtroom bench and gavel, the setting where New Mexico's Supreme Court sanctioned a lawyer for filing a brief containing ChatGPT-fabricated testimony.
A courtroom bench and gavel, the setting where New Mexico's Supreme Court sanctioned a lawyer for filing a brief containing ChatGPT-fabricated testimony.

New Mexico's Supreme Court has fined a defense attorney $5,000 and held him in contempt after finding that the appellate brief he filed for a client convicted of murder quoted police officers and eyewitnesses that ChatGPT had invented.

Key takeaways

  • The New Mexico Supreme Court fined Santa Fe attorney Stephen Aarons $5,000 on Wednesday, held him in contempt, and said it will refer him to the state attorney disciplinary board.
  • The court found the brief contained false testimony from wholly fabricated witnesses, including invented statements that the shooter wore dark pants and a white shirt.
  • Aarons said he fed a machine-generated trial transcript to ChatGPT expecting a bulletproof summary and did not grasp how far the tool could hallucinate.

The sanction was first reported by Reuters and set out in a court filing issued on Wednesday. The justices said Aarons failed to verify the factual claims and legal authority in his AI-generated brief, and that he had demonstrated a lack of remorse and a lack of concern for his client.

What the fabricated brief actually contained

Aarons was handling the appeal of Oscar Renee Sandoval, who was convicted last year of murdering the mother of his children and sentenced to life in prison. The primary brief filed on Sandoval's behalf did not merely cite cases that do not exist, the failure mode that has become routine in AI sanctions rulings. It attributed testimony to witnesses who never took the stand and to police officers who never gave the statements described.

The court had ordered Aarons last month to explain how the invented material reached the filing, singling out what it described as fictional statements about the shooter's clothing. At an August 21 hearing, Aarons told the justices he had fed a computer-generated transcript and other case materials into a chatbot and assumed it would return a reliable digest of the trial.

Justice C. Shannon Bacon appeared unconvinced that a practicing attorney could have missed the well-documented risk. She asked Aarons whether he watched the news, listened to the radio, or read anything about what was happening in the world, noting that lawyers relying on AI hallucinations had become an above-the-fold story every single day, according to an account of the hearing.

In a statement to Reuters, Aarons said he was remorseful but hoped the disciplinary board would treat the filing as an honest mistake, and called it a lesson for every professional relying on a powerful but sometimes unstable technology. OpenAI, which builds ChatGPT, did not immediately comment.

Why this case escalates the pattern

Dozens of American lawyers have now been disciplined for filings in which a large language model manufactured citations or misquoted binding law. Lawyers representing MyPillow founder Mike Lindell were fined over AI-generated misquotes, and a federal judge last year criticized two firms for a brief riddled with false and misleading citations.

Fabricated witness testimony in a criminal appeal is a different order of error. A wrong citation can be caught by opposing counsel or a clerk checking a reporter volume. Invented trial testimony corrupts the factual record a court is being asked to review, in a case where the defendant is serving a life sentence. Sandoval's appeal remains pending and was reassigned on September 2 to Kim Chavez Cook, a New Mexico public defender, who declined to comment. The district attorney's office for Dona Ana County also declined.

The episode fits a broader pattern of AI output entering public records faster than anyone verifies it. Research this year found that a large share of newly published web pages show signs of AI authorship, and courts are now discovering that professional filings are no exception.

What happens next

The disciplinary referral is the part with real consequences. A $5,000 fine is survivable for a private practitioner; a suspension is not. Several state and federal courts have already begun requiring certifications that filings were checked against primary sources, and rulings like this one give those rules teeth. For firms deploying AI tools, the operative lesson is narrow and unglamorous: summarization of a transcript is a retrieval task, and retrieval still has to be checked line by line against the record.

FAQ

How much was the New Mexico lawyer fined for using ChatGPT?

The New Mexico Supreme Court fined Stephen Aarons $5,000 and held him in contempt. The court also said it would refer him to the state attorney disciplinary board for a separate investigation, which could carry heavier professional penalties.

What did ChatGPT fabricate in the brief?

The court said the brief contained false testimony from wholly fabricated witnesses, plus invented police testimony and false details about the shooter's clothing and appearance. Aarons had asked the chatbot to summarize a computer-generated trial transcript.

Does the murder conviction still stand?

Yes. Oscar Renee Sandoval's appeal is still pending and was reassigned on September 2 to a New Mexico public defender. The sanction applies to his former lawyer's conduct, not to the merits of the appeal itself.

How do you feel about this article?

SJ

Discussion

Sign in to post
Loading...

Related articles

Washington's $1 ChatGPT Deal Expires. Its Replacement Bills by the Token.
Tech & Business

Washington's $1 ChatGPT Deal Expires. Its Replacement Bills by the Token.

OpenAI's new 27-month GSA deal waives a $15 per-seat license and halves token rates, but replaces a $1-a-year flat fee with metered billing from October 1.

Seung Jung4 days ago
Altman Rules Out an OpenAI IPO in 2026 Even With a Confidential S-1 on File
Tech & Business

Altman Rules Out an OpenAI IPO in 2026 Even With a Confidential S-1 on File

OpenAI will not complete an initial public offering this year, chief executive Sam Altman said in an interview with Fortune editor in chief Alyson Shontell, ans...

Seung Jung4 days ago
OpenAI Says Its Own Models Helped Tape Out Jalapeño in Nine Months
Tech & Business

OpenAI Says Its Own Models Helped Tape Out Jalapeño in Nine Months

AI-generated kernels beat OpenAI expert-written versions by up to 1.8x, as Jalapeño posts its first InferenceX benchmark results.

Seung Jung22 days ago
Judge Voids Pentagon Supply Chain Risk Label on Anthropic as Unlawful Retaliation
Tech & Business

Judge Voids Pentagon Supply Chain Risk Label on Anthropic as Unlawful Retaliation

A 59-page order found the Pentagon retaliated against Anthropic for criticizing the government, violating the First and Fifth Amendments.

Seung Jung20 days ago
Starcloud Raised $250M for Orbital AI Compute. Its Hardest Problem Is Booking a Rocket.
Tech & Business

Starcloud Raised $250M for Orbital AI Compute. Its Hardest Problem Is Booking a Rocket.

The Series A extension doubles Starcloud to a $2.3B valuation, but Falcon 9 winds down in 2028 and Starship has yet to prove rapid reuse.

Seung Jung25 days ago
Feds Warn AI-Written Exploits Are Already Hitting Critical Infrastructure Controllers
Tech & Business

Feds Warn AI-Written Exploits Are Already Hitting Critical Infrastructure Controllers

Five US agencies say attackers are using AI coding assistants and the open source snap7 library to build tools that read and write Siemens S7 controller memory.

Seung Jung28 days ago