New Mexico court fines lawyer for ChatGPT's fabricated witnesses
A state supreme court just treated an AI-hallucinated appellate brief as what it is — a professional betrayal, not a tech mishap. Plus: Moonshot AI puts a revenue number on its IPO year.
The New Mexico Supreme Court has held veteran criminal defense lawyer Stephen Aarons in direct contempt for filing a murder-appeal brief stuffed with "false testimony from wholly fabricated witnesses." Aarons admitted he fed a computer-generated trial transcript and case documents into ChatGPT, signed the output without verifying a word of it, and never told his client. The court fined him $5,000 to the State Bar's Client Protection Fund, struck his briefs from the record, barred him from appearing before it pending a disciplinary investigation, and ordered a new lawyer for the defendant — a man serving a life sentence whose appeal now restarts from zero.
The fabricated names are what make the order read like a warning label. The brief quoted an Officer Michelle Amarillo and other witnesses who never testified, invented threats from real people, and misdescribed the actual case law it cited. At last month's hearing, Justice C. Shannon Bacon dismantled the claim that Aarons didn't know models hallucinate — her 13-year-old nephew and 75-year-old stepmother both know, she said, so "either you buried your head in the sand … or you took a gamble." Aarons, a lawyer of over 40 years, told Ars Technica he is "remorseful but hopeful" the disciplinary board sees "an honest mistake." The court's answer was blunt: the duty to verify applies whether the work came from an AI tool, a law student, or a first-year associate — and the lawyer who signs the brief owns every word.
Fake-citation cases have drawn fines and referrals since 2023; this one escalates to direct contempt because the harm wasn't abstract — a real client, a life sentence, an appeal torched. What to watch: whether New Mexico adopts the AI-use certificate Aarons himself suggested for every filing, and whether the disciplinary board adds to the fine.
Moonshot AI is targeting $2 billion in annualized revenue for 2026, according to Bloomberg — the growth number behind this morning's confidential Hong Kong IPO filing. We covered the filing earlier today — Moonshot AI files for Hong Kong IPO at $50B — days after Anthropic calls it out — and the revenue target completes the picture: TechCrunch's report leans on OpenRouter's public usage tracker as evidence Kimi's paid adoption is real. Context check, though: $2 billion is small next to OpenAI's roughly $40 billion run-rate and Anthropic's $65 billion annualized. The subtext matters more than the figure — Moonshot needs its IPO priced off commercial momentum, not off the distillation scandal Anthropic pinned on it this week. Pre-IPO revenue claims deserve roadshow-deck scrutiny, but if Kimi's usage keeps compounding, the reported $50 billion valuation stops looking like fiction.
If a lawyer's AI-generated filing can burn a life-sentence appeal, should courts require a signed AI-verification certificate on every brief? Tell us in the comments.
