Lawyer Fined $5,000 After AI Invented Witnesses in Murder Appeal

Introduction

Generative AI can produce polished legal prose, but fluency is not evidence. New Mexico’s Supreme Court has fined lawyer Stephen Aarons $5,000 and held him in contempt after an appeal brief in a murder case contained fabricated material generated with the help of ChatGPT.

According to the reported filing, the brief included testimony from witnesses who did not exist. It also contained false testimony about police evidence and inaccurate descriptions of the shooter’s clothing and appearance. These were not merely errors in a private draft: they appeared in a document submitted as part of an appeal challenging a murder conviction.

Key points

• The court focused on a failure to verify. The New Mexico Supreme Court said Aarons did not check the factual claims or legal authority in the AI-generated brief. That failure led to the fine and contempt finding.
• The lawyer misunderstood the tool’s limits. During an August hearing, Aarons acknowledged using ChatGPT and said he expected it to create a “bulletproof summary” of the trial. The episode illustrates how excessive confidence in a model can become a professional failure, even when the user did not intend to mislead the court.
• AI hallucinations are now a recurring courtroom problem. Justice C. Shannon Bacon questioned how Aarons could have missed the risks, noting that lawyers’ reliance on AI hallucinations has become a widely reported issue. Other proceedings have involved fabricated citations, incorrect quotations, and misleading legal authorities submitted by attorneys or law firms.

Why it matters

Legal filings require more than coherent language. Every witness, quotation, factual assertion, case citation, and procedural detail must be traceable to the record or to an authoritative legal source. A generative model predicts plausible text; it does not independently know whether a witness appeared at trial or whether a quotation exists in a judgment. When information is missing, it may fill the gap with details that sound precise but have no basis in the case.

That makes AI more suitable for limited support tasks—such as organizing notes, suggesting a structure, or helping locate material—than for serving as an unchecked source of facts. Human review must include comparison with the original record, verification of citations, and confirmation that every material assertion is supported. In criminal proceedings, the stakes are especially high because fabricated claims can affect the rights of the accused and impose additional burdens on courts and opposing parties.

The sanction also clarifies accountability. Using an AI tool is not automatically the same as violating professional rules, but submitting unverified output to a court remains the responsibility of the lawyer who signs and files it. Law firms and courts may therefore need clearer policies, source tracking, review checklists, and pre-filing audits. The practical lesson is simple: AI may accelerate drafting, but it cannot replace legal judgment or factual verification.

Source: The Verge AI