Authorized to Draft Is Not Authorized to File

The order is short on ambiguity. Stephen Aarons, counsel of record in a criminal appeal before the Supreme Court of New Mexico, used ChatGPT to help prepare an appellate brief. The brief that reached the court contained, as the court's own Dispositional Order of Direct Contempt states, false testimony from wholly fabricated witnesses and misrepresented legal authority. Aarons acknowledged as much, and he went further: he admitted he did not verify the factual claims and legal authority the tool produced before he signed the brief and filed it.

The court's response was not a warning. It held Aarons in direct contempt, struck all briefing filed in the matter, replaced him as counsel, referred him for professional discipline, barred him from appearing before the court pending that process, and imposed a five thousand dollar sanction.

Capability to prepare, not authority to file

Nothing in the order suggests that using an AI tool to help draft a brief was itself the violation. What the order treats as the failure is narrower and more precise: Aarons signed and filed the brief without independently checking whether what it said was true. Preparing a document and being entitled to submit it as an accurate, verified representation to a court are two different authorities, and this order is direct evidence that a professional obligation to verify sits between them regardless of which tool, or which person, produced the first draft. An attorney's authority to use a drafting aid does not, on its own, ever extend into authority to skip the check that turns a draft into a filing.

That distinction matters beyond this one case because the same shape recurs whenever a system, human or automated, is authorized to prepare a consequential output and nothing separate confirms that output before it takes effect. Here the separation was never automated away: it was a professional's own duty, and the order's own account is that the duty was not performed, not that no mechanism existed to perform it.

What the record establishes, and what it does not

This record is built from three excerpts of the court's own order, mirrored in full text by Justia, since a court hosted written-order URL was not the route through which the order text was captured. The Supreme Court of New Mexico's own recordings-of-oral-arguments listing independently confirms the case number and its show cause hearing, and Reuters reported the same five thousand dollar sanction and the same ChatGPT attribution independently. Between them, an official docket confirmation and independent journalism corroborate the order's own account.

What this record does not establish is also worth stating plainly. It does not identify which specific witnesses were fabricated or which specific authority was misrepresented beyond the order's own quoted language. It does not establish how the fabrications were first discovered, or by whom. It does not state the eventual outcome of the discipline referral against Aarons, which the order refers onward rather than resolving itself. Those facts are recorded here as unknown, not inferred from what a similar case might suggest.

Where this sits against what Moona Intelligence has already written

The gap between authority to prepare a change and authority to enact it recurs across this desk's own reporting, most often as an agent delegated both halves of a workflow that was designed to keep them separate. This record extends that same authority gap to a different shape: not an agent holding both roles, but a human professional who occupied the separate, supposedly independent role and did not perform the check that role existed to provide. The gap is the same. The actor who failed to keep it closed is not.