The Estate Fight That Turned Into an AI Sanctions Case

Joanne Couvrette's lawsuit against members of the Wisnovsky family began as an estate and trust dispute in Oregon federal court. That type of case is messy enough on its own: relatives fighting over property, competing versions of what a decedent intended, and a record thick with amendments, transfers, and accusations of self-dealing. None of that is unusual. What made Couvrette v. Wisnovsky unusual was that the summary-judgment phase turned into a full demonstration of how badly AI-assisted legal briefing can go when nobody takes responsibility for checking the authorities before filing.

By the time Magistrate Judge Mark Clarke was finished with it, the case had generated one of the most severe AI-related courtroom outcomes on the site so far: repeated sanctions findings, a fee award of $94,704.38 assessed directly against the lawyers, and dismissal with prejudice.

Fifteen Fake Cases and Eight Fabricated Quotations

The core factual problem was not subtle. As the court later summarized, plaintiffs' summary-judgment briefing contained 15 non-existent or otherwise fake case citations and eight fabricated quotations across three separate briefs. This was not a single bad string cite buried in a footnote. It was a pattern of made-up authority embedded in the papers that were supposed to win the case on the merits.

The court's orders make clear why this one escalated so far. Defendants' lawyers noticed the fake cases and fabricated quotations and put plaintiffs' counsel on notice. That should have forced an immediate stop, a real citation audit, and a blunt explanation to the court. Instead, the record as described by Judge Clarke showed superficial or inadequate correction attempts and a continuing failure to deal honestly with the scale of the problem.

That detail matters more than the raw citation count. Federal judges are capable of distinguishing between an error that is caught quickly and owned promptly, and a bad filing that keeps getting defended after everyone in the room knows it is contaminated. Couvrette drifted into the second category.

December Was the First Hammer

On December 12, 2025, Judge Clarke issued the first major sanctions order. The court found repeated violations tied to the fake authorities and local-rule problems in the summary-judgment briefing. Plaintiffs' briefing was struck. The court also set in motion additional scrutiny of local counsel Timothy Murphy, who had allowed his name to remain on the case while failing to meaningfully participate in a way the court considered compatible with his obligations as local counsel.

That point is easy to miss if the story gets reduced to "lawyers used AI and got caught." One of the court's themes was that professional obligations do not disappear just because work is being done by someone else in the chain. Lead counsel cannot treat AI as a black-box intern and sign the paper anyway. Local counsel cannot treat the matter as somebody else's problem while still lending their name and admission status to the filing. The rules of responsibility remain older and stricter than the software.

The December order already put the plaintiffs' side in a hole deep enough to make climbing out difficult. The court was not merely irritated about sloppy legal writing. It concluded that the briefing process had crossed the line into sanctionable conduct.

March Was the Accounting

The March 23, 2026 fee order turned that conduct into a number. After reviewing the defendants' amended bill of costs and attorney fees, Judge Clarke awarded $94,704.38 as the reasonable attorney fees and costs directly resulting from the plaintiffs' defective summary-judgment submissions. The court assigned 85% of that amount, $80,498.72, to pro hac vice counsel Stephen Brigandi and 15%, $14,205.66, to Murphy.

Even in a profession accustomed to sanctions stories, that is a serious number. It is not a symbolic penalty designed to sting without changing behavior. It is a six-figure consequences package in all but rounding error, tied directly to the opposing side's cost of dealing with fake citations, fabricated quotations, and the procedural chaos that followed.

The March 23 order is also notable for the way it handled Murphy. Judge Clarke did not just order payment. He also required that Murphy attach the opinion to future pro hac vice motions he sponsors as local counsel in the District of Oregon. That is a practical reputational penalty. It means the sanctions order does not stay tucked away in Westlaw for specialists to find later. It follows him into future applications.

Then the Case Died

A week later, on March 30, 2026, the court dismissed the plaintiffs' claims with prejudice. At that point the fake-citation problem had fully merged with the merits posture of the case. The court had already concluded the summary-judgment briefing was unusable. Dismissal closed the loop. Whatever the plaintiffs thought they were buying when they hired lawyers and trusted their papers to survive judicial scrutiny, this was the opposite outcome.

That is one reason this story hits harder than the more familiar AI-lawyering incidents. Many citation-hallucination cases end with a fine, a lecture, and a bruised professional reputation. Couvrette ended with the underlying litigation collapsing. The lawyers did not just embarrass themselves. They lost the case in a way that became inseparable from the briefing misconduct.

Why the Court Took It So Seriously

The legal profession has spent the past few years pretending there is still some unresolved mystery about how to use generative AI in litigation safely. There is not. The rule is boring and already settled: if you cite a case, you are responsible for making sure the case exists and says what you claim it says. If you quote an opinion, you are responsible for confirming the words are actually in the opinion. If an opponent tells you your authorities are fake, the next move is not "keep improvising." The next move is "stop filing things until a human has checked everything."

Couvrette reads like what happens when that rule is ignored at multiple stages. Fake authority entered the briefs. Opposing counsel flagged it. The response did not satisfy the court. Local counsel failed to act like local counsel. Lead counsel could not distance himself from the product filed under his name. The judge then used the tools federal courts already have: strike the briefing, shift the fees, and end the case.

None of that required a new anti-AI doctrine. It required ordinary Rule 11 principles applied to a newer failure mode.

The Local Counsel Problem

One of the most useful parts of this case is how plainly it rejects the idea that local counsel is a decorative role. In multi-jurisdiction litigation, local counsel sometimes gets treated as a procedural adapter: the lawyer who helps out-of-state counsel get admitted and then quietly recedes into the wallpaper. Judge Clarke's orders make clear that this is not how the court sees it. If your name is on the filing and your status is helping the case stay in court, you are not exempt from responsibility when the papers contain fabricated authority.

That point will matter beyond Oregon. A lot of AI-assisted litigation work is already being split across firms, contract lawyers, staff attorneys, and outsourced drafting processes. Couvrette is a reminder that the court will not spend much time sorting out who pressed which button if the final filing is false.

Why This One Belongs High on the List

There are now plenty of stories where a lawyer filed a few fake AI citations and paid for it. This one stands out because the court treated the problem as sustained litigation misconduct with case-ending consequences. The fee award was large. The dismissal was final. The court's description of the record left no room for the softer version of events in which everyone was trying their best and the chatbot merely got overexcited.

The useful distinction is between a filing mistake and a process collapse. A filing mistake can be corrected. A process collapse keeps producing bad paper after the warning lights are already on. Couvrette was the second kind. That is why it belongs on the site, and why any firm still treating AI citation verification as an optional cleanup step should read the orders with a calculator open.