“Because the problem with lawyers relying on AI hallucinations is an above-the-fold story every single day. So either you have buried your head in the sand … and that’s an intentional choice to be uninformed, or you took a gamble, and neither of those are consistent with the code of conduct.”

— New Mexico Supreme Court Justice C. Shannon Bacon

Search our catalog of “live” and “on-demand” continuing legal education (CLE) webinars for practical courses covering Artificial Intelligence that can help you avoid similar issues. Credit available in multiple states.

For attorneys in all practice areas, the promise of generative artificial intelligence (Gen AI) has been framed as a revolution in efficiency. But as a recent New Mexico Supreme Court disciplinary proceeding vividly illustrates, offloading cognitive legal work to an AI chatbot without rigorous verification is a fast track to professional catastrophe.

In a stark reminder of the ethical obligations governing technological competence, the New Mexico Supreme Court recently sanctioned veteran Santa Fe criminal defense attorney Stephen Aarons for submitting an appellate brief riddled with AI-generated hallucinations. Aarons’ reliance on ChatGPT – resulting in disciplinary actions – should serve as a cautionary tale for the broader legal community regarding the collision of artificial intelligence tools and the Model Rules of Professional Conduct.

The Factual Background: A High-Stakes Appeal

Aarons, an attorney with more than 40 years of experience, was representing a 60-year-old defendant who had been convicted of murdering his wife. The original criminal trial took place in the Las Cruces area (within New Mexico’s Third Judicial District Court of Doña Ana County).

Faced with voluminous trial transcripts and complex court documents, Aarons opted to use ChatGPT to synthesize the material and draft his brief. He mistakenly believed that ChatGPT’s large language model (LLM) would generate a “bulletproof” summary of the record.

The AI Misstep: Fabricating the Record

As technologists and legal ethicists have repeatedly warned, LLMs are not built to retrieve concrete facts; they are built to predict statistically probable text based on their training data. When asked to summarize dense, factual transcripts without strict guidelines, chatbots frequently “hallucinate.” If a human produced similar “hallucinations” in their work-product, they would be called “errors”, “mis-statements”, “or lies”.

In Aarons’ case, the AI completely fabricated portions of the trial record. The appellate brief submitted to the New Mexico’s highest court included references to entirely non-existent witnesses. Even more damaging to the integrity of the appeal, the AI generated fictional testimony and falsely attributed it to at least one real witness who had actually testified during the Las Cruces murder trial.

Aarons submitted the brief without adequately verifying the AI’s output against the official court transcripts—a fundamental breach of an appellate lawyer’s duty of diligence.

The Supreme Court’s Rebuke and Sanctions

When the fabricated testimony was discovered by the New Mexico Supreme Court, the court’s response was swift and decisive. During the ensuing disciplinary hearing, Aarons admitted to his “stupidity”, but the justices firmly rejected technological ignorance as a viable defense.

Justice C. Shannon Bacon led a blistering critique from the bench, questioning how a practicing lawyer could be unaware of the well-documented risks of AI hallucinations. “Do you watch the news? Do you listen to the radio? Do you read anything about what’s going on in the world?” Bacon asked, pointing out that AI errors in legal filings are heavily publicized.

The sanctions handed down by the court were comprehensive and immediate:

• Contempt of Court: The Supreme Court formally found Aarons in contempt for filing the fabricated brief.

• Financial Penalty: Aarons was ordered to pay a $5,000 fine to a client protection fund within a 30-day window.

• Removal from Representation: The court removed Aarons from the Las Cruces murder appeal entirely, transferring the 60-year-old defendant’s case to a public defender to ensure competent ongoing representation.

• Disciplinary Referral: The incident was referred to the state’s disciplinary board for further review, potentially jeopardizing Aarons’ license to practice law.

Aarons attempted to mitigate the sanctions by noting he had recently completed Continuing Legal Education (CLE) courses on AI use, suggesting that mandatory tech-focused CLEs could prevent similar errors among peers. The court remained unmoved, emphasizing that the burden of competence rests squarely on the individual practitioner.

Ethical Implications for the Legal Profession

This incident forces a critical re-examination of how the legal profession interprets its ethical duties in the digital age. Under ABA Model Rule 1.1 (Competence), lawyers must maintain the legal knowledge, skill, thoroughness, and preparation necessary for representation. Comment 8 explicitly states that lawyers must keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology.

Furthermore, this case heavily implicates Rule 3.3 (Candor toward the Tribunal), which strictly prohibits a lawyer from making a false statement of fact or law to a tribunal. When an attorney signs a pleading or brief, they are certifying its accuracy to the court. The origin of the text—whether drafted by a junior associate, a paralegal, or an AI chatbot—does not shield the signing attorney from their Rule 3.3 obligations.

The New Mexico Supreme Court’s ruling echoes the fallout from the infamous 2023 Mata v. Avianca case in the Southern District of New York, where attorneys were sanctioned for submitting fake judicial opinions generated by ChatGPT. The fact that similar disciplinary crises are still unfolding years later indicates a persistent and dangerous disconnect between the rapid adoption of legal tech and basic ethical oversight.

Key Takeaways for Practitioners

Artificial intelligence is an undeniably powerful tool for legal research, document review, and drafting routine correspondence. However, it cannot replace the duties of diligence and competence owed to a client or the candor owed to the court.

For legal professionals, the Aarons case provides three crucial takeaways:

1. Verification is Non-Negotiable: Every citation, legal rule, and factual assertion generated by an AI must be manually verified against the primary source record before filing.

2. Ignorance is Not a Defense: Claiming a lack of understanding regarding how an LLM operates will not mitigate disciplinary sanctions. If you use a tool in your practice, you are strictly responsible for understanding its limitations.

3. The Record is Sacred: In appellate law, the trial record is the absolute universe of facts. Fabricating that record, even inadvertently through a technological proxy, fundamentally compromises the justice system and will trigger severe judicial reprisal.

As generative AI becomes deeply embedded in legal workflows, the New Mexico Supreme Court has set a firm boundary: innovation must never come at the expense of professional integrity.