A New Mexico defense attorney is $5,000 poorer after submitting an appellate brief full of AI-generated fiction, including entirely fake witnesses and fabricated police testimony. The state Supreme Court didn’t just reject the filing. It held Stephen Aarons in contempt.

The Wednesday ruling marks one of the first times a state’s highest court has formally sanctioned a lawyer for AI hallucinations in a criminal case. And it wasn’t a close call. According to the court’s filing, Aarons’ brief “contained false testimony from wholly fabricated witnesses” along with bogus statements attributed to law enforcement. All of it generated by ChatGPT. None of it verified.

Aarons was representing a client appealing a murder conviction. That’s high stakes. The kind of case where you’d expect extra scrutiny of every claim. Instead, the court found he failed to “verify the factual claims and legal authority in his AI-generated brief.”

His defense? “I didn’t know that AI could hallucinate facts,” Aarons told reporters, according to Ars Technica.

This isn’t new, but courts are getting less patient

Lawyers getting burned by ChatGPT isn’t novel anymore. The most famous case is probably Mata v. Avianca, where a New York attorney cited six completely fake cases in a federal filing. That lawyer got sanctioned too, and the case became a cautionary tale in every CLE course on AI tools.

But the New Mexico case is worse in a specific way. Aarons didn’t just cite nonexistent case law. He invented witness testimony. In a murder appeal. That crosses from careless research into fabricating evidence, even if the fabrication was outsourced to a chatbot.

The $5,000 fine is notable but not devastating. It’s the contempt finding that stings. That goes on your record. It signals to other courts that you didn’t just make a mistake, you failed a basic professional duty.

What this means for lawyers using AI

The legal profession has been slow to figure out how to regulate AI use. Some courts now require explicit disclosures when lawyers use generative AI for briefs. Others have issued standing orders warning attorneys that they remain responsible for everything they file, regardless of what tools they used to draft it.

New Mexico’s Supreme Court is now firmly in the “we will punish you” camp. The ruling makes clear that using AI doesn’t lower the bar for factual accuracy. It raises it. If you’re going to use these tools, you’d better triple-check the output.

That’s a reasonable standard. Lawyers already have ethical obligations to verify facts and avoid frivolous filings. AI doesn’t change that. But the technology makes it easier to generate plausible-sounding nonsense at scale, which means the consequences when someone skips verification are worse.

The “I didn’t know” defense doesn’t work

Aarons’ claim that he didn’t know AI could hallucinate is hard to take seriously at this point. By 2026, ChatGPT’s tendency to confidently invent information has been widely reported for years. Every major legal publication has run warnings. Bar associations have issued guidance. If you’re using the tool in a professional capacity, you’re expected to understand its limitations.

The court clearly didn’t buy it. The contempt finding suggests they viewed this as willful ignorance at best.

For criminal defense attorneys specifically, this case is a warning shot. Your client’s freedom might depend on the brief you file. If you’re going to use AI to draft it, you need to treat every sentence like it might be fiction. Because it might be.

The New Mexico Supreme Court’s ruling won’t be the last word on AI in legal practice. But it’s a clear signal that courts are done being patient with lawyers who treat ChatGPT like a research assistant without doing the actual research.