US lawyer sanctioned for submitting ChatGPT-fabricated witnesses in murder appeal

Oscar Renee Sandoval was convicted and sentenced to life in prison; the integrity of his murder appeal was compromised by fabricated evidence in his defense brief.
The problem with lawyers relying on AI hallucinations is an above-the-fold story every single day.
A New Mexico Supreme Court justice expressed frustration that defense attorney Stephen Aarons appeared unaware of well-documented AI risks.
Mark

So a lawyer just... trusted ChatGPT to write part of a murder appeal without checking it?

Mimi

He fed the trial transcript into ChatGPT and asked it to summarize the proceedings. He thought it would produce a solid summary. Instead, it invented witnesses and testimony that never happened.

Luke

Did he test it first? Did he run a smaller task through it to see if it was reliable?

Mimi

There's no indication he did. He seems to have assumed that if you gave it real case materials, it would work accurately.

Mark

And the court was... not sympathetic?

Mimi

A justice asked him if he watched the news, because AI hallucinations in legal filings have been a major story for months. The court said he showed a lack of remorse.

Luke

But here's the thing—how many other lawyers are doing this right now without getting caught? This one got caught because the other side or the court noticed the fabrications.

Mimi

That's a real question. This is the first case I've seen where AI invented actual witness testimony in a criminal appeal, but there could be others.

Mark

What happens to his client now?

Mimi

The appeal was reassigned to a public defender. The client is serving life for murder. His original defense lawyer just compromised his appeal with made-up evidence.

Luke

And we don't know if those fabrications actually hurt the appeal's chances, or if they were just embarrassing. The appeal is still pending.

Mark

So this lawyer gets fined and referred to a disciplinary board, but his client is the one who really pays.

Mimi

Exactly. Aarons said it was an honest mistake. The court disagreed.

  • A murder appeal brief filed on behalf of a man serving life in prison contained fabricated witnesses and invented police testimony that never appeared anywhere in the actual trial record.
  • The New Mexico Supreme Court discovered the fabrications during review, finding that the attorney had fed case materials into ChatGPT and submitted its output without any verification whatsoever.
  • Justice C. Shannon Bacon expressed open exasperation at Aarons's claimed ignorance of AI hallucinations, noting that the phenomenon had been front-page news for years.
  • Aarons was fined $5,000, held in contempt, and referred to the state disciplinary board — while his client's compromised appeal was quietly reassigned to a public defender.
  • The case marks what legal observers see as a new escalation: not merely fabricated citations, but wholly invented witness testimony introduced into a criminal proceeding where a human life hangs in the balance.

In Santa Fe, a defense attorney's misplaced trust in an artificial intelligence tool has placed both his career and his client's freedom in jeopardy. Stephen Aarons submitted a murder appeal brief containing fabricated witnesses and invented police testimony generated by ChatGPT — details he never verified against the actual record. The New Mexico Supreme Court, confronting what may be a new threshold in AI-assisted legal error, sanctioned Aarons and referred him for disciplinary review, while his client, Oscar Renee Sandoval, continues to serve a life sentence with his appeal now in other hands. The case asks a question courts are increasingly forcing into the open: when professionals delegate judgment to machines they do not fully understand, who bears the cost of that ignorance?

Stephen Aarons, a defense lawyer in Santa Fe, fed trial transcripts into ChatGPT expecting a reliable summary for his client's murder appeal. What the system produced instead was a brief populated with fabricated police testimony and witnesses who never existed — including invented descriptions of a shooter's clothing that appeared nowhere in the actual record.

The New Mexico Supreme Court uncovered the fabrications while reviewing the appeal of Oscar Renee Sandoval, convicted of murder and sentenced to life in prison. The court fined Aarons $5,000 and held him in contempt, noting not only his failure to verify the filing's accuracy but his apparent lack of remorse and concern for his client — a man whose only avenue for relief had been undermined by his own counsel.

At an August hearing, Aarons admitted he had not understood how AI systems can invent facts entirely — what researchers call hallucination. Justice C. Shannon Bacon responded with visible frustration, pointing out that AI hallucinations in legal filings had become a daily headline. Aarons later described the episode as an honest mistake, though courts have grown less patient with that explanation.

The case sits at a new threshold. While dozens of lawyers have already been sanctioned for briefs containing fabricated case citations, the introduction of wholly invented witness testimony in a criminal appeal represents a graver breach — one that implicates not just professional standards but the integrity of the justice system itself. Aarons has been referred to the state disciplinary board. Sandoval's appeal has been reassigned to a public defender and remains pending. OpenAI did not respond to questions about how its system generated such specific false details, or what safeguards exist to prevent similar incidents.

Stephen Aarons, a defense lawyer in Santa Fe, made a decision that would cost him $5,000 and his professional standing. He fed trial transcripts and case materials into ChatGPT, expecting the artificial intelligence system to produce what he called a "bulletproof summary" for his client's murder appeal. What he got instead was a brief containing entirely fabricated police testimony and witnesses who never existed—details so specific they included descriptions of a shooter wearing dark pants and a white shirt, statements that appeared nowhere in the actual record.

The New Mexico Supreme Court discovered the fabrication when reviewing Aarons's filing in the appeal of Oscar Renee Sandoval, who had been convicted and sentenced to life in prison for murder. On Wednesday, the court fined Aarons and held him in contempt for failing to verify the accuracy of what he had submitted. The justices went further, noting that Aarons had "demonstrated a lack of remorse and a lack of concern for his client"—a particularly sharp rebuke given that Sandoval's only hope for relief now rested on an appeal that had been compromised by its own defense counsel.

When the court ordered Aarons to explain himself at an August hearing, he acknowledged that he had not fully grasped how AI systems can invent facts wholesale, a phenomenon researchers call "hallucination." Justice C. Shannon Bacon responded with visible exasperation, asking whether Aarons watched the news or read anything about the world, noting that AI hallucinations in legal filings had become "an above-the-fold story every single day." Aarons later told Reuters he was remorseful but characterized the error as an honest mistake born of not understanding the technology's limitations.

The case is not isolated. Courts across the country have begun disciplining lawyers for submitting AI-generated documents without proper review. Dozens have been sanctioned for briefs containing fabricated case citations or misquoted law. But Aarons's filing appears to represent a new threshold—the introduction of wholly invented witness testimony in a criminal appeal, where the stakes involve a person's freedom and the integrity of the judicial process itself.

The New Mexico Supreme Court has referred Aarons to the state's attorney disciplinary board for investigation. His client's appeal has been reassigned to Kim Chavez Cook, a public defender, who declined to comment. The appeal remains pending. Sandoval, who pleaded not guilty to murdering the mother of his children, is serving a life sentence based on a conviction that now must be defended by someone other than the lawyer who initially took up his cause. OpenAI did not respond to requests for comment on how its system came to generate such specific false details, or what safeguards, if any, exist to prevent similar incidents.

I am remorseful but hopeful that the disciplinary board takes into account it was an honest mistake. It is a lesson learned for all professionals who rely upon this powerful but sometimes unstable technology.
— Stephen Aarons, defense attorney, in statement to Reuters
Counsel, do you watch the news? Do you listen to the radio? Do you read anything about what's going on in the world? Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day.
— Justice C. Shannon Bacon, New Mexico Supreme Court, at August 21 hearing
Möchten Sie die ganze Geschichte? Das Original lesen bei Al Jazeera ↗
Kontakt FAQ