Judge Mark D. Clarke called the fabrications "a notorious outlier in both degree and volume," handed down $110,000, and dismissed an elder-abuse case with prejudice for a client who never touched the briefs.
In a Medford, Oregon courtroom this spring, U.S. Magistrate Judge Mark D. Clarke wrote that the case in front of him was "a notorious outlier in both degree and volume." Then he handed down the largest AI-hallucination sanction in the history of American courts. The number was $110,000. The dismissed claim behind it was worth $12 million.
The attorneys were Stephen Brigandi and Tim Murphy. They represented Joanne Couvrette in an elder-abuse case in the U.S. District Court for the District of Oregon. The case was the kind plaintiff lawyers fight to keep alive. It had a sympathetic client. It had a seven-figure ask. It died with prejudice, which means it can never be refiled, and it died not on the merits but on the conduct of the lawyers who filed the briefs.
Judges have been issuing AI sanctions for two years now. The running tally of these cases now numbers in the hundreds. Most run a few thousand dollars. A lawyer cites a case that does not exist, the court catches it, the lawyer apologizes, and a modest fine follows. The pattern had become almost routine. Clarke saw something different on his docket. He counted the fabrications. He measured the volume. And he reached for the word "outlier" because nothing else fit.
The volume is what set this apart. This was not one phantom citation slipped into one brief. It was a body of work. Fake cases. Fake quotations attributed to real cases. Made-up propositions of law dressed up to look like settled doctrine. The court did not find a careless lawyer who leaned on a tool once. It found filings shot through with material the machine had invented and the humans had signed.
Clarke did not hide his reasoning behind procedural language. He wrote that "if there was ever an 'appropriate case' to grant terminating sanctions for the misuse of artificial intelligence, this is it." That is a judge telling the bar that he knows the line he is drawing. Terminating sanctions are the heaviest weapon a court holds. They do not punish a filing. They end the case. Clarke aimed that weapon at the plaintiff's $12 million claim and pulled the trigger.
Think about what that costs a client. Joanne Couvrette did not write the briefs. She did not prompt the model. She hired counsel, the way every litigant does, and trusted that the words filed under her name would be real. The terminating sanction fell on her case all the same. The elder-abuse claim was extinguished. The path to a $12 million recovery closed. The lawyers' conduct became the client's loss, which is the quiet cruelty of a dismissal with prejudice.
The $110,000 is the part that will travel. It is the headline number, and it should be. But the dollar figure is the smaller half of the story. The dismissal is the larger half. A sanction is money. A terminating sanction is the case itself. Clarke chose both. He fined the lawyers and he killed the claim, and he wrote his order in a way that makes clear he understood the weight of doing both at once.
Trial drama, nuclear verdicts, and the plaintiff-firm tactics behind them. Court-reporter prose, no consultant filler. Read by litigation leaders at F500 legal departments and national carriers. Free.
The phrase "in both degree and volume" is doing precise work. Degree is how bad each error was. Volume is how many there were. A judge willing to separate those two axes is a judge who has read the filings closely enough to grade them. Clarke did not wave at "AI errors" as a category. He took the measure of these particular errors, in this particular case, and found them off the chart on both axes at once.
The misuse here is a kind every trial lawyer now recognizes. The tool is fast. It is confident. It produces citations that look exactly like citations. It produces quotations that read exactly like quotations. The only thing it does not reliably produce is truth. A lawyer in a hurry, facing a deadline, asks the model for support and gets back a brief that looks finished. The trap is that "looks finished" and "is real" are no longer the same thing, and the gap between them is where careers go to die.
Brigandi and Murphy walked into that gap. The record does not show two lawyers who set out to deceive the court. It shows something more ordinary and more dangerous. Lawyers who trusted a machine, did not check its work, and signed their names to its output. The signature is the act that matters. When a lawyer signs a filing, the lawyer certifies that the contents are true and warranted. The machine cannot make that certification. Only the human can, and Brigandi and Murphy did, on filings full of things that were not so.
Clarke's order arrives at a moment when courts across the country are still calibrating. Some judges issue warnings. Some impose small fines. Some require continuing-education hours on AI. The bench has not converged on a number or a rule. Into that uncertainty, an Oregon magistrate dropped the ceiling. Not a thousand dollars. Not ten thousand. One hundred ten thousand, plus the death of the underlying claim. Every judge weighing an AI sanction now has a marker on the high end, and a colleague who explained, in plain words, why he set it there.
The "outlier" framing cuts two ways, and Clarke surely knew it. An outlier is, by definition, the exception. He is saying this conduct was worse than the run of cases. But an outlier is also a data point that pulls the whole distribution. The next judge who sees a pile of fabricated citations will have Clarke's order in hand. The next lawyer who is tempted to file unchecked AI output will have a number in mind. Outliers do not stay outliers for long once they are written into an order and indexed in a database. They become the case other judges cite.
There is a coverage wrinkle worth noting for anyone watching this from the carrier side. The reporting on Clarke's ruling broke in May 2026, in OPB and Fortune, even though the order itself is docketed earlier. The lag shows how these stories surface. The order sits quietly in a federal docket for months. Then it breaks into the trade press, and suddenly the number is everywhere, and the bar is talking about it. The $110,000 was always real. It just took a news cycle to make it loud.
The machine was wrong. The lawyers signed anyway. The judge counted the lies. The claim is dead.
The Executive Briefing is six questions. It shows you exactly where the gaps are.
Take the Executive Briefing →Trial drama, nuclear verdicts, and the plaintiff-firm tactics behind them. Court-reporter prose, no consultant filler. Read by litigation leaders at F500 legal departments and national carriers. Free.