Lawyer Fined $5,000 for AI-Hallucinated Witnesses in Murder Appeal

New Mexico's Supreme Court fined attorney Stephen Aarons $5,000 and held him in contempt after his AI-generated appellate brief, filed on behalf of a client convicted of murder, contained testimony from wholly fabricated witnesses and false details about the shooter's clothing and appearance. The court ruled that Aarons failed to verify the factual claims and legal authority in his brief. Justice C. Shannon Bacon raised the failure at an August hearing, questioning why Aarons appeared unaware of the known risks of using AI for legal work.

What happened

Detail	Fact
Attorney	Stephen Aarons
Court	New Mexico Supreme Court
Fine	$5,000
Sanction type	Contempt of court
Case type	Murder conviction appeal
Hearing date	August (year not specified in source)
Filing date	Wednesday (per Reuters report)

Stephen Aarons submitted an appeal brief for a client challenging a murder conviction. According to a court filing reported by Reuters, the brief was AI-generated and included testimony from witnesses who do not exist, as well as false claims about the shooter’s clothing and appearance. The New Mexico Supreme Court found that Aarons made no meaningful effort to check whether the facts or legal citations in the document were real.

The court fined him $5,000 and held him in contempt. Justice C. Shannon Bacon specifically questioned Aarons at an August hearing about why he appeared oblivious to well-documented risks that AI tools can fabricate sources, citations, and testimony.

Why it matters

This is not an isolated incident. Courts across the US have now sanctioned multiple attorneys for submitting AI-generated documents containing invented case law or fictitious facts. What makes this case particularly serious is the stakes: a murder conviction appeal, where fabricated witness testimony could have directly misled the court about what happened at the scene.

The pattern is consistent. A lawyer uses an AI tool, likely a general-purpose chatbot such as ChatGPT, trusts the output, and submits it without cross-checking against real court records or case files. AI language models (tools that generate text by predicting likely word sequences) do not “know” whether a witness existed. They produce plausible-sounding text, and in a legal brief, plausible-sounding is dangerous.

For businesses that rely on AI-generated content, whether for contracts, marketing copy, or regulatory documents, the lesson is the same: AI output requires human verification before it goes anywhere official.

Our take

At Lumien, we use AI tools daily across AI integration projects and content workflows. The productivity gains are real. But this case is a useful reminder that AI confidence and AI accuracy are completely different things. A model that invents a witness does not flag it as uncertain; it just writes it as fact.

Professionals in any field who put AI-generated content in front of a regulator, a judge, or a paying client carry the same responsibility as they always did. The tool does not absorb that liability. Aarons is $5,000 poorer and in contempt of the state’s highest court, and his client’s appeal was presumably not helped either.

If you are reading about AI hallucinations for the first time, check our broader AI news coverage for context on how often this kind of failure shows up in production environments. It is frequent enough to take seriously.

What to do about it

1. Treat every AI output as a draft, not a final document, especially for anything that goes to a third party.
2. Cross-check all named people, citations, case numbers, and statistics against primary sources before submission.
3. Establish a written verification step in your workflow so it cannot be skipped under deadline pressure.
4. If AI tools are used in client-facing work, disclose it and confirm your review process covers factual accuracy, not just grammar.

The fine is small. The reputational and legal exposure for your client is not.

Source: The Verge · AI