A federal court rules judges keep immunity even if they let AI write the ruling

A federal court has ruled, for what appears to be the first time, that a judge who allegedly let AI write a ruling from start to finish is still absolutely immune from being sued over it. U.S. District Judge Gloria Navarro of the District of Nevada dismissed Phillips v. Parladé on August 17, 2026, without ever deciding whether the underlying accusation — that a state family-court judge delegated a decision entirely to AI — was true. Her reasoning: even if it happened exactly as alleged, choosing to use AI would still be the judge's own judicial act, and judicial acts carry absolute immunity regardless of how badly they're performed.

The plaintiff, T. Matthew Phillips, a California-licensed attorney representing himself, sued Nevada family-court judge Mari Parladé over a ruling in his own underlying state case. According to Above the Law's account of the complaint, his theory rested on a few concrete details: Parladé appeared to be reading from a prepared script during proceedings, stumbled over a word, and at one point misspoke "appearance of any impartiality" for the standard legal phrase "appearance of impropriety." Phillips argued those details showed the ruling wasn't really hers — and that judicial immunity, the doctrine shielding judges from civil suits over their official acts, shouldn't cover a decision no human judge actually made.

It's worth being precise about what Navarro's order does and doesn't establish, because the underlying facts are contested and unresolved. Above the Law reports this was at least Phillips's tenth federal lawsuit against state court officials, and none of the prior nine succeeded — context that matters for how much weight the specific factual allegations deserve on their own. Navarro's order never says Parladé used AI; it says the immunity analysis comes out the same way even assuming, for the sake of argument, that she did.

To get there, Navarro applied the ordinary four-factor test federal courts use to decide whether an act is "judicial in nature" and therefore immune: whether it's a normal judicial function, whether it happened in the judge's own chambers, whether it centered on a case actually pending before her, and whether it arose from a party's dealings with the judge in her official capacity. Issuing a ruling in a pending family-law case checked every box — the same test any ordinary contested ruling would pass, with nothing added or subtracted for the AI allegation.

“Judges enjoy absolute immunity from civil liability, even if their action was in error, done maliciously, or in excess of their authority.” — from Judge Navarro's order, as quoted by Reason's Volokh Conspiracy

That line states a doctrine far older than AI. The U.S. Supreme Court recognized absolute judicial immunity in the 1870s (Bradley v. Fisher) and reaffirmed it in 1978 in Stump v. Sparkman, a case where a judge approved a mother's petition to sterilize her teenage daughter without a hearing, notice, or any legal basis — and was still held immune, because approving the petition was a judicial act, however badly performed. What's new here is a court applying that same century-old doctrine, seemingly for the first time on the record, to an allegation that a judge handed a ruling to a machine rather than simply ruling badly herself. Navarro's opinion treats "used AI to decide" as functionally identical to "decided badly" or "decided with an undisclosed bias" — categories immunity has always covered — rather than carving out a new, AI-specific hole in the doctrine.

(Immunity from a civil damages suit isn't the only accountability channel a judge faces — it just closes off this particular one. A state judicial-conduct commission could still investigate the same underlying allegation on its own authority, using a lower bar than a federal civil suit requires.) Nothing in Navarro's order forecloses that route, and nothing in it says Parladé actually did what Phillips claims. What it settles is narrower and, for the AI industry, more durable: as machine-assisted drafting spreads through courts, agencies, and other places that issue binding decisions, the officials using it are not automatically stripped of the legal protections that came with the job before AI existed. The next case to test that line will need a plaintiff whose factual allegations survive first contact with a court — this one never got that far.