Lawyer Fined $5,000 — Used AI-Fabricated Witnesses in Murder Appeal

The New Mexico Supreme Court has fined lawyer Stephen Aarons $5,000 for submitting a legal brief that included AI-fabricated witnesses and phony police testimony. According to a report from The Verge, the fine was issued Wednesday as the court held Aarons in contempt for using the wholly invented material in an appeal for his client's murder conviction.

This incident is another stark example of the pitfalls of integrating generative AI into high-stakes professional work without a fundamental understanding of the technology. The tools are not databases of fact; they are sophisticated text predictors. The distinction proved costly for Aarons.

A "Bulletproof" Brief Riddled with Holes

The lawyer’s goal was to create a "bulletproof summary" for the appeal, as he told the court, according to The Guardian. Instead, the brief submitted to the state's highest court was compromised by hallucinations from OpenAI's ChatGPT. The AI model invented police testimony and even fabricated entire witnesses, weaving them into the legal argument as if they were real.

Both sources confirm the core of the issue: the brief contained entirely made-up information. This wasn't a case of misinterpreting a legal precedent. It was the submission of pure fiction as fact in a proceeding where a person's liberty was on the line. The court's action underscores that the ultimate responsibility for the accuracy of a legal filing rests squarely with the lawyer who signs it, not the software used to help draft it.

A Failure of Verification, Not Technology

The court’s punishment was not for using AI, but for the gross failure of professional duty. The Verge reports that Aarons was held in contempt specifically for his failure to "verify the factual claims" made in the document. This is the central point. The legal system, like many professions, relies on its practitioners to be guarantors of fact. Introducing an unreliable narrator like an un-checked LLM into that process breaks the chain of trust.

Together, these reports point to a recurring and dangerous pattern. Professionals, pressed for time and lured by the promise of AI-driven efficiency, are deploying these tools without adequate safeguards or training. They are treating generative AI as a research assistant rather than what it is: a powerful but flawed content generator that requires rigorous human oversight. This case indicates that courts have little patience for a learning curve when fundamental legal ethics are breached.