The Ninth Circuit Just Drew The Line: It Is Not The AI Hallucination, It Is The Cover-Up

Posted June 4, 2026

A judge's gavel resting on a desk in a courtroom, representing the Ninth Circuit sanctions against lawyers who filed AI-hallucinated case citations

For three years, the AI-lawyer disaster genre has followed a comfortable script. An attorney asks a chatbot for case law, the chatbot invents some, the attorney files it, a judge catches it, everyone gets a news cycle and a modest fine, and the profession collectively promises to be more careful. The Ninth Circuit Court of Appeals just rewrote the ending. In a published opinion this week, the court sanctioned two California lawyers, Mike Singh Sethi and William Rounds, over appellate briefs stuffed with nonexistent cases, misattributed quotations, and misrepresented authority. And the court went out of its way to explain what actually earned them the punishment. It was not the hallucinations. It was the months they spent pretending the hallucinations were typos.

Read that again, because it is the new rule of the road. The most AI-skeptical federal appeals court moment to date is not about a machine making things up. Courts have made their peace with the fact that chatbots fabricate. What the Ninth Circuit refused to absorb was lawyers fabricating about the fabrication.

What Actually Happened

The briefs at issue contained the now-familiar fingerprints of generative AI run unsupervised: citations to cases that do not exist anywhere in any reporter, real cases quoted as saying things they never said, and authority characterized in ways the actual opinions do not support. When the problems surfaced, the attorneys had the standard fork in the road. Option one, own it: admit the AI use, apologize, correct the record. Option two, deny everything and hope the panel loses interest.

They chose option two. The errors were blamed on typographical mistakes. The suggestion that AI produced them was repeatedly denied. A federal appellate panel, which reads briefs for a living, was asked to believe that a string of fully formed phantom cases with plausible-looking citations materialized through typos. The panel did not believe it, and the opinion's tone makes clear how much the denial cost.

The Bill

The sanctions package is worth itemizing, because it is the heaviest combination an appeals court has handed down in one of these cases:

That last item is the innovation, and it deserves attention. The court did not just punish past conduct. It built a supervision regime. For the next two years, these lawyers practice with a scarlet annotation, required to flag their AI use in everything they file. It is the legal profession's version of an ignition interlock device, and it establishes a template any court in the country can now copy.

And the panel said the quiet part in writing: lesser sanctions may have been warranted if the attorneys had disclosed their AI use and apologized. The court even spelled out the protocol it expects, that a lawyer who learns of any error in a filing, including a generative AI hallucination, should immediately alert the court and opposing counsel and disclose its source. That sentence is now effectively the candor standard for AI mistakes in the federal appellate system. Confession gets you a fine. Concealment gets you a suspension and a two-year leash.

The Pattern Courts Are Done Tolerating

What makes this opinion land harder than the dozens of trial-level sanctions before it is the institutional weight behind it. District judges have been improvising responses to hallucinated filings since 2023, with penalties ranging from wrist slaps to six figures. The result was a patchwork that let lawyers treat AI citations as a low-grade gamble. A published appeals court opinion changes the math. It is precedent, it is citable, and it announces that the era of the typo defense is over.

The timing is not an accident either. The same week, Florida's Supreme Court posted a new rule aimed squarely at AI hallucinations in court filings, formalizing what judges there expect when a machine-generated citation slips through. Two of the most influential judicial systems in the country reached for structural fixes within days of each other. That is not coincidence. That is a profession's immune system finally responding to an infection it spent three years documenting case by case.

Why Lawyers Keep Doing This

The uncomfortable question underneath every one of these stories is why highly credentialed professionals keep walking into the same propeller. The answer is not stupidity. It is that generative AI is engineered to sound finished. A hallucinated case does not look like a draft or a guess. It arrives formatted, confident, and wrapped in the cadence of real legal writing. Verification feels redundant precisely because the output performs reliability. That performance is the product.

Add the economics, where legal research is expensive, deadlines are brutal, and a chatbot produces in nine seconds what an associate produces in nine hours, and the temptation writes itself. The machine does not bill, does not sleep, and does not push back. It also does not know what a fact is. Every sanctioned lawyer in this genre made the same trade, swapping verification for velocity, and most of them would have survived the swap if they had simply told the truth when caught.

The Disaster Is Now A Candor Problem

That is the real shift to take from this week. The AI-in-court disaster has matured. Phase one was novelty, the first wave of fake citations and the first incredulous judges. Phase two was volume, hallucinated filings piling up faster than courts could shame them away. Phase three, starting now, is enforcement infrastructure: published precedent, disclosure regimes, standing rules, and a clear price list. The hallucination costs you embarrassment and a fine. The cover-up costs you your seat at the table.

The machines, for their part, have not changed at all. They will keep generating authoritative fiction for as long as anyone asks. The legal system has simply stopped pretending to be surprised, and started treating every unverified AI citation as what it always was, a loaded filing waiting to go off in someone's career.