The Phantom Witnesses: When a Murder Appeal Meets AI Hallucination

When you outsource a mundane administrative task to an AI, the worst-case scenario is usually a typo or a formatting error. But when you outsource a murder appeal to a chatbot, the consequences can derail the justice system entirely.

Recently, the New Mexico Supreme Court confronted a stark example of professional negligence fueled by artificial intelligence. Attorney Stephen D. Aarons was representing a man serving a life sentence for shooting his wife. In his effort to draft an appeal, Aarons turned to ChatGPT, expecting the tool to generate a "bulletproof summary of proceedings."

Instead, the AI did what large language models frequently do when they lack context: it improvised. The chatbot fabricated two witnesses, "Danny Stanton" and "Linda Stanton," and generated detailed, entirely fictional testimonies regarding threats and the shooter's appearance. Aarons submitted the brief without verifying the existence of these individuals.

When caught, the lawyer's defense during a disciplinary hearing was essentially a plea of technological ignorance. He called it an "honest mistake" and blamed his own "stupidity," noting he had heard that doctors used AI for research and assumed it was safe for legal briefs.

The judges were unsparing in their response. They dismantled the notion that a professional in 2024 could simply be unaware of AI's tendency to invent facts—a phenomenon widely known as "hallucination." As one judge pointed out, lawyers failing their clients through AI-generated errors is practically a daily news story. "My 13-year-old nephew knows about hallucinations," the judge remarked, framing Aarons' actions not as a simple mistake, but as a willful choice to remain uninformed or a reckless gamble with a client's life.

The fallout was swift and severe. Aarons was found in direct contempt of court, fined $5,000, removed from the murder case, and referred to the Disciplinary Board. Perhaps most troubling of all, Aarons admitted he hadn't directly informed his incarcerated, Spanish-speaking client about the ChatGPT debacle, telling the family only that there was a "problem with the brief."

This incident is more than just a cautionary tale about a lazy shortcut. It strikes at the heart of professional accountability in the AI era. Generative AI tools are designed to predict the next plausible word in a sentence, not to act as arbiters of truth. They can mimic the tone of legal authority perfectly while delivering pure fiction. As AI integration becomes inevitable across industries, professionals must remember one immutable rule: you can outsource the drafting, but you can never outsource the liability.

Key Points

• A defense attorney used ChatGPT to draft an appeal for a client convicted of murder, resulting in a brief filled with fabricated witnesses and fake testimony.
• The lawyer claimed ignorance about AI 'hallucinations,' an excuse the judges flatly rejected as willful blindness.
• The attorney faced severe sanctions, including a $5,000 fine, removal from the case, and a referral for disciplinary action.
• The case serves as a stark reminder that while AI can assist in drafting, humans remain entirely liable for verifying the facts.

Why It Matters

As AI tools become mainstream, professionals can no longer use 'technological ignorance' as a shield. This case sets a clear boundary: the convenience of AI does not override the fundamental duty of human verification.

Sources:

• Lawyer Cites ChatGPT-Invented Fake Witnesses in Murder Appeal — 404 Media