TL;DR

New Mexico’s highest court has fined an attorney $5,000, held him in contempt and referred him for disciplinary investigation after his appeal brief in a murder case relied on witnesses and police evidence that did not exist. He had fed the trial record to ChatGPT expecting a reliable summary. This is a meaningful escalation: previous sanctions have concerned invented case law, not invented testimony.

What went into the brief

Santa Fe attorney Stephen Aarons had taken on an appeal for Oscar Renee Sandoval, sentenced to life last year for murdering his children’s mother. The brief he filed, according to the court, “contained false testimony from wholly fabricated witnesses”. Among the invented material were “fictional statements that the shooter was wearing dark pants and a white shirt”.

At a hearing on 21 August, Aarons explained that he had handed the model a machine-generated transcript along with other case papers, assuming what came back would be “a bulletproof summary”. He told Reuters he had not grasped how readily the technology hallucinates, and described the episode as “an honest mistake” and “a lesson learned for all professionals who rely upon this powerful but sometimes unstable technology”.

The bench was unsympathetic. Justice C. Shannon Bacon asked whether he watched the news or read anything about the world, adding: “Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day.” The panel also found that he had “demonstrated a lack of remorse and a lack of concern for his client”. The appeal has since been reassigned to a public defender.

Why this one is different

Dozens of lawyers have now been disciplined for briefs containing fabricated citations or misquoted law. Those errors are embarrassing but self-limiting: a citation either resolves or it does not, and opposing counsel checks. Invented testimony, attributed to witnesses who never existed, is harder to catch and goes to the facts a court must weigh.

Looking forward

For UK firms the exposure runs through the SRA’s competence duties rather than contempt powers, but the failure mode is identical, and it compounds the warning that feeding legal papers to public AI may waive privilege. Summarisation is the task most solicitors have already delegated, and it is precisely where a model will fill a gap it cannot see.