A federal magistrate judge in Oregon has ordered two lawyers to pay roughly $110,000 in fines and opposing-counsel fees after they filed court briefs — the written arguments lawyers submit to a judge — containing 15 entirely fictional court cases and eight made-up quotations. The penalty, handed down in a family winery dispute called Couvrette v. Wisnovsky, is believed to be the largest U.S. sanction — a formal penalty imposed by a court — so far in a growing category of legal misconduct: lawyers using artificial intelligence to draft filings and then submitting what the software invents without checking whether any of it is real.
The case was dismissed with prejudice, meaning it cannot be refiled. Magistrate Judge Mark D. Clarke wrote that “in the quickly expanding universe of cases involving sanctions for the misuse of artificial intelligence, this case is a notorious outlier in both degree and volume.”
What an AI “hallucination” actually is
When people talk about AI hallucinations, they are describing a specific failure mode of tools like ChatGPT, Claude, or Google’s Gemini. These programs — known as generative AI, because they generate new text rather than retrieving stored facts — work by predicting the most likely next word in a sentence. They are extraordinarily good at producing fluent, confident writing. They are not designed to verify whether what they are writing is true.
So when a lawyer asks one of these tools for cases supporting a legal argument, the AI may produce something that looks exactly like a real court citation — a plausible case name, a plausible court, a plausible year, even a plausible-sounding quotation from the judge. The problem is that none of it exists. The case was never filed. The judge never wrote those words. The quotation was assembled from statistical patterns, not from law.
That is a hallucination: confident, polished output with no connection to reality.
Why this case stood out
The Oregon dispute involved a long-running feud over a family-owned winery. According to Judge Clarke’s December 12, 2025 opinion, San Diego attorney Stephen Brigandi — appearing in Oregon under a temporary admission known as pro hac vice, a procedure that lets a lawyer licensed in one state appear in a single case in another — filed three briefs that included 15 nonexistent cases and eight fabricated quotations. His Portland-based local counsel, Tim Murphy, signed off without independently checking the work.
According to the New York Times, the bogus citations kept arriving even after opposing lawyers had pointed out earlier ones. Two appeared in a January 2025 filing, then seven more in April, then 16 more in May.
Judge Clarke was unsparing. “Rather than a correction, Mr. Brigandi attempted a cover-up,” he wrote. “He failed at both. There is no evidence that Mr. Brigandi took any steps to verify the veracity of the legal arguments.” The judge faulted both attorneys for being “not adequately forthcoming, candid or apologetic.”
The total penalty, split across two rulings, comes to about $15,500 in disciplinary fines plus roughly $94,705 in opposing-counsel fees. Brigandi shoulders the larger share — about $80,500 in fees against him, plus a portion of the disciplinary fines. Murphy owes about $14,206. In a statement, Murphy said: “I did not write, review, research, sign or submit the briefs with AI citations. I am more than very sorry for my involvement in this matter.”
“Behind each of those cases is a client who paid a lawyer and trusted the system.”
Alexandra Smyth, General Counsel, LexisNexis Legal & ProfessionalHow common is this?
Damien Charlotin, a Cambridge-trained lawyer and senior research fellow at HEC Paris, maintains a public database of court decisions worldwide that address AI hallucinations in legal filings. As of late May 2026, the database listed 1,494 such decisions, with 1,034 from U.S. courts.
Those numbers come with important caveats. Charlotin only tracks decisions where a judge addressed the AI issue substantively, and most state-court rulings never appear in the legal databases researchers can search. The 1,494 figure is best understood as a floor, not a complete count — and it cannot tell us whether the underlying behavior is accelerating, plateauing, or simply being detected more often.
What is clearer is that courts are increasingly willing to punish the conduct when they find it. In March 2026, the Sixth Circuit Court of Appeals sanctioned two Knoxville attorneys $15,000 each — plus their opponents’ legal fees and double court costs — for briefs containing “over two dozen fake citations and misrepresentations of fact.” Days later, an Oregon appeals court fined a Salem attorney $10,000 for citations the court said had been “contrived from thin air.” Around the same time, the Nebraska Supreme Court indefinitely suspended a lawyer for what reports describe as dozens of defective citations in a single brief.
None of this means AI is unreliable for legal work as a category. Specialized legal research tools from companies like LexisNexis and Thomson Reuters are designed differently from general chatbots, though a 2024 Stanford study found that even those purpose-built tools hallucinate between 17% and 33% of the time. The pattern in nearly every sanctions case is not that a lawyer used AI — it is that a lawyer used AI and then filed the result without checking it.
What this means for you
If you hire a lawyer for anything — a divorce, a business dispute, an immigration matter, an estate, or a personal injury claim — you have no easy way to tell whether the brief filed on your behalf cites real cases. You are paying for the lawyer’s judgment, but you are also paying for their verification. The cases above suggest those two things are not always happening together.
Before your next consultation, or at your next check-in with an attorney you already work with, consider asking these three questions. Each tracks a failure pattern that judges have specifically punished.
- 1“Do you use generative AI to draft or research my filings, and if so, which tools?” You are not asking this to catch them out. AI use is now common and, when done carefully, legitimate. You are asking to establish that the lawyer is honest about their workflow. A vague or defensive answer is itself useful information.
- 2“How do you personally verify every case citation and quotation before something is filed in my matter?” The test in every sanctions ruling has been independent verification against established legal databases like Westlaw or Lexis — not asking the AI whether its own output is real. That is the standard. Ask whether your lawyer meets it.
- 3“If you discover a mistake in a filing already submitted in my case, what do you do?” In nearly every six-figure sanction, the aggravating factor was not the initial error — it was the delay, the deflection, or the attempted cover-up afterward. Immediate disclosure to the court is what separates a warning from a career-altering penalty.
A lawyer who answers these three questions thoughtfully is not necessarily a great lawyer. But a lawyer who cannot answer them, or who bristles at being asked, is telling you something worth knowing.
The Oregon ruling will not be the last word. Bar associations in more than 35 states have now issued AI guidance, but the rules vary widely and there is no uniform federal standard. Watch for two developments that would meaningfully change the picture: a sanction against a partner at a large national firm — so far, an analysis by the Stanford Cyberlaw Center found that 90% of the firms involved in AI-tainted filings are solo or small practices of under 25 attorneys — or a federal rule requiring lawyers to disclose AI use in filings. Either would move the country from a patchwork of one-off rulings into a unified standard. Until then, much of the burden of catching the problem still falls, awkwardly, on clients themselves.
