AI-Generated Falsehoods Bring $5,000 Fine for Lawyer in New Mexico Murder Appeal

A New Mexico lawyer has been fined $5,000 and held in contempt after submitting an appeal brief containing fabricated witnesses, invented testimony and false details generated with the help of ChatGPT.

The case is a warning about the legal risks of treating generative artificial intelligence as a reliable source of facts. The New Mexico Supreme Court said lawyer Stephen Aarons failed to verify both the factual claims and the legal authorities in an AI-generated filing.

The case was reported on September 11, 2026. The brief concerned a client’s murder conviction and included fictional evidence about the case.

What the New Mexico court found

According to the court filing, Aarons’s brief contained:

• Wholly fabricated witnesses
• False testimony attributed to people who did not exist
• Incorrect testimony concerning the shooter’s clothing and appearance
• Unverified legal and factual claims produced by an AI system

The court held Aarons in contempt and imposed the $5,000 fine. The sanction followed an August hearing at which he acknowledged using ChatGPT and said he had expected the system to produce what he described as a “bulletproof summary” of the trial.

AI systems can produce fluent, plausible text while inventing people, events, quotations and authorities. In a legal filing, persuasive wording does not substitute for evidence or verification.

Why the case matters

The New Mexico matter is more serious than a routine citation error. A fabricated legal citation can undermine an argument, but invented witnesses and testimony can distort the factual record in a criminal case.

Courts rely on lawyers to distinguish among:

• Evidence admitted at trial
• Statements made by actual witnesses
• Arguments advanced by the parties
• Legal authorities that genuinely exist
• Assertions that still require investigation or proof

An AI-generated summary can blur those categories, particularly when a lawyer assumes that a coherent narrative reflects the underlying record. The danger is greatest when the tool is asked to summarize a long trial transcript or construct an appellate argument without being checked against the original documents.

Lawyers remain responsible for filings submitted in their name, regardless of whether a human assistant or an AI tool helped prepare them.

The warning from Justice C. Shannon Bacon

During the August hearing, Justice C. Shannon Bacon questioned how Aarons could have been unaware of the risks associated with AI-generated legal material. Her remarks reflected the extent to which fabricated citations and other AI errors had already become a widely publicized problem.

Aarons said he was remorseful and hoped a disciplinary board would regard the incident as an honest mistake. The court’s action nevertheless demonstrates that unfamiliarity with an AI tool is unlikely to excuse a lawyer from checking its output.

What lawyers can learn from the case

AI tools may assist with tasks such as organizing notes, generating possible headings or suggesting questions for further research. They should not be treated as an independent source of testimony, precedent or procedural history.

Before submitting an AI-assisted brief, a lawyer should:

1. Check every citation against the original authority.
2. Read the cited decision rather than relying on an AI-generated quotation.
3. Compare factual summaries with the record, transcripts and exhibits.
4. Confirm that every witness, quotation and procedural event actually exists.
5. Review the final filing as the responsible author, not as a passive editor of machine-generated text.

These steps protect the client’s legal position and the court’s ability to decide a case on an accurate record.

A broader pattern of AI-related legal errors

The New Mexico case follows other reported incidents in which lawyers submitted briefs containing fabricated or inaccurate legal material. The source material also refers to a previous case involving law firms whose filing included numerous false, inaccurate or misleading citations and quotations, as well as sanctions involving lawyers for Mike Lindell over AI-generated misquotes and fake citations.

Those examples point to a recurring failure: the problem is not necessarily the use of AI itself, but the removal of human verification from a professional process that depends on primary documents.

A lawyer who uses an AI system for drafting still has to validate the result line by line. Without that validation, the technology can turn an ordinary research mistake into a filing containing an entirely fictional account of the case.

Source

• The Verge, published 2026-09-11: https://www.theverge.com/ai-artificial-intelligence/994207/chatgpt-new-mexico-lawyer-fined-murder-appeal