AI generated fake testimony lawyer sanction triggers contempt

The New Mexico Supreme Court on September 11, 2026 held defense attorney Stephen Aarons in direct contempt after his appellate brief, produced with ChatGPT, contained "false testimony from wholly fabricated witnesses" and mis-cited legal authority. The court noted that Aarons admitted he never verified the AI-generated content before filing and failed to inform his client of the errors. This AI generated fake testimony lawyer sanction underscores that reliance on large language models does not replace the attorney's duty to check facts.

Technical roots of the hallucination

ChatGPT’s model, released in early 2025, excels at natural-language generation but retains stochastic sampling that can produce hallucinations—plausible-sounding statements that lack factual basis. In this case the model was fed a computer-generated transcript of a murder trial and invented witnesses such as Officer Michelle Amarillo and Manal Al-Jibury. Because the model does not have real-time access to court records, any fabricated detail must be caught by the user before submission.

Court’s reasoning and professional-conduct implications

Justice C. Shannon Bacon emphasized that the lawyer violated multiple provisions of the New Mexico Bar’s code of conduct, including candor to the tribunal and competence. The opinion made clear that AI hallucinations are a known risk and do not excuse a lawyer from independent verification. The ruling aligns with emerging ethical guidance that requires attorneys to treat AI as a drafting aid, not a source of truth.

AI generated fake testimony lawyer sanction implications

The sanction sends a clear market signal. Law firms must now build verification pipelines that pair AI-drafted text with human fact-checking and citation tools. Bar associations are likely to issue advisory opinions mandating audit trails for AI contributions. Firms that ignore these steps risk disciplinary action, financial penalties, and damage to client trust.

Operational fallout for the case

Aarons was barred from appearing before the Supreme Court pending disciplinary board findings and fined $5,000, payable to the State Bar of New Mexico Client Protection Fund. The court also struck all prior briefs from the record and ordered the public defender’s office to appoint new counsel for Oscar Renee Sandoval. These actions illustrate how a single AI error can cascade into procedural delays, higher litigation costs, and potential appeals.

Broader impact on legal-tech adoption

While AI tools are increasingly used for document review, brief drafting, and case summarization, this incident highlights the need for hybrid workflows. Companies may invest in automated citation checkers, provenance logs, and layered human review. The episode also accelerates interest in specialized legal LLMs trained on verified case law, reducing reliance on generic models that lack domain-specific grounding.

Regulatory outlook

State bar disciplinary boards are expected to update ethical guidelines, potentially codifying explicit verification requirements for AI-assisted work. Federal regulators may extend consumer-protection scrutiny to legal-tech vendors that make unverified accuracy claims. Legislative bodies could consider audit-trail mandates similar to provisions in the EU AI Act for high-risk systems.

Incentives, risks, and what comes next

Lawyers are motivated to adopt AI for speed and cost savings, but the Aarons case shows that shortcuts can backfire with severe professional consequences. Risks include reputational harm, client mistrust, and costly sanctions. Moving forward, we expect tighter oversight, mandatory training on AI limitations, and the emergence of certification programs for AI-assisted legal drafting. Stakeholders should watch for new bar opinions and possible state legislation that formalizes AI verification standards.

What lawyers should do now

  1. Implement dual-review processes – Pair AI-generated drafts with manual fact-checking before filing.
  2. Maintain source logs – Record which portions of a brief originated from AI and retain the original prompts.
  3. Stay informed on AI risks – Follow updates from professional bodies and technical standards organizations.
  4. Use trusted tools – Rely on platforms that provide citation verification and provenance tracking.

Related research

Recent work on mitigating hallucinations in large language models is documented in a paper hosted on Papers with Code, illustrating techniques that could eventually reduce incidents like Aarons’ brief. See the code alongside the paper.

Source

The full court order and case details are available at Ars Technica.

Related coverage

Explore more on this topic