Oregon Judge Drops $110K on Two Lawyers for AI Hallucinations. The Details Are Worse Than the Number.

A federal magistrate judge in Oregon just handed down a $110,000 sanctions order against two lawyers who filed briefs with 15 nonexistent cases and eight fabricated quotations. The opinion (issued Dec. 12 by U.S. Magistrate Judge Mark D. Clarke, District of Oregon, and reported this week) calls the case “a notorious outlier in both degree and volume.” It is the largest AI hallucination sanction ever issued by an Oregon federal judge, and it should be required reading for every managing partner in the country.

The headline is the dollar figure. The substance is the conduct around the errors, because Judge Clarke did not just sanction the AI misuse. He sanctioned the cover-up.

The facts

The underlying case was an Oregon family winery dispute. Joanne Couvrette sued her two brothers for control of their family winery, alleging elder abuse and wrongful enrichment, with $12 million on the line. She was represented by San Diego attorney Stephen Brigandi, admitted pro hac vice, and because Brigandi was not Oregon-licensed, Portland landlord attorney Tim Murphy served a procedural role as local counsel.

During briefing on cross-motions for summary judgment, Couvrette’s side filed three briefs containing 15 references to cases that did not exist and eight quotations that were simply made up. Defendants’ counsel flagged the errors.

What happened next is the part that drove the number up. According to Clarke’s opinion, Brigandi “attempted a cover-up” by removing the false information and refiling his argument. The court found that he took no steps to verify the legal arguments in the amended briefing, which the judge said still contained misstatements of law.

“Plaintiffs and their counsel have not been adequately forthcoming, candid or apologetic about their conduct,” Clarke wrote. “If there was ever an ‘appropriate case’ to grant terminating sanctions for the misuse of artificial intelligence, this is it.”

The judge dismissed the case with prejudice. He ordered Brigandi to pay $80,000 in attorney fees and roughly $15,000 in fines. Murphy was ordered to pay $14,000 in additional fees. The case was gone. The client lost the $12 million claim, not on the merits but on sanctions.

The breakdown matters. Clarke also noted that Couvrette herself shared responsibility. Defendants described her as “a serial self-represented litigator who was likely the person responsible for using artificial intelligence” to draft the briefs, and her lawyers provided no contrary evidence. Murphy told reporters he “did not write, review, research, sign or submit the briefs with AI citations.” That is not a defense. He signed as local counsel. The $14,000 bill followed.

Why this one is different from the other hallucination cases

There are now hundreds of reported AI hallucination sanctions in U.S. courts. The database at damiencharlotin.com has been tracking them in near real time. Most of those cases end with a four- or five-figure fine, a mandatory CLE requirement, and a public scolding. This one is different for three reasons.

Volume. Fifteen nonexistent cases and eight fabricated quotations across three briefs is not a slipped citation. It is an entire body of fake law. The opinion describes it as “notorious” specifically because of the scale.

The cover-up. Courts are learning to separate the first mistake from the second one. The first mistake is filing bad cites. The second mistake is refiling without genuinely auditing the work and without apologizing. Clarke’s opinion treats the cover-up as an independent aggravating factor, which is where the $80,000 in fee-shifting comes from.

Terminating sanctions. Case dismissed with prejudice. The client with the $12 million claim is out. That is the big-picture consequence that malpractice carriers and clients should actually care about. The fine is the smallest part of the damage.

The attorney liability landscape is now clear. If a lawyer files AI-generated work, fails to cite-check, gets caught, and then tries to paper over the problem without a full audit and full candor, the court has both the tools and the appetite to end the case.

The local counsel problem

Murphy’s situation is the one that should scare every partner who agrees to serve as local counsel for an out-of-state firm. He said, credibly by all accounts, that he did not write, review, research, sign, or submit the offending briefs. The court sanctioned him $14,000 anyway.

The message to every local-counsel arrangement in the country: your name on the case is enough. If you are letting lead counsel file briefs without your review, you are underwriting their AI practices. Either review the work, get paid for the review, and document your independent judgment, or decline the engagement.

The days of a cheap local-counsel role (sign the pro hac motion, collect a modest fee, stay out of the substance) are over. The new floor is: if you are appearing in the caption, you are on the hook for AI hallucinations in the filings, whether you wrote them or not.

The pro se client problem

The other piece buried in Clarke’s opinion is the finding that the client herself was likely the one generating the AI content. Couvrette, described as “a serial self-represented litigator,” apparently drafted or helped draft the briefs using AI, and her lawyers filed them with insufficient independent verification.

This is a live risk pattern. Clients with strong views, technical comfort, and access to ChatGPT or Claude will generate “helpful” drafts and send them over. The lawyer’s job (and the lawyer’s liability) is to treat that input as raw material, not finished work. When a client hands you a brief, run the citations, regenerate the analysis from primary sources, and document what you verified. The old rule was: do not let the client write your briefs. The new rule is: especially do not let the client’s AI write your briefs.

What to do now

If you run a law firm or a legal department, tighten five things this week.

1. Mandatory cite-check with a live reporter. Every citation in every filing gets pulled and read against the source. Not read in an AI summary. Read in Westlaw, Lexis, or Bloomberg Law. If your AI tool cannot show you the actual case in a primary database, assume the case does not exist.

2. Second-lawyer signoff on any brief that used AI assistance. The lawyer who used the tool is not the right person to audit the output. A colleague with no sunk cost in the draft should read every citation before filing.

3. A hallucination response protocol. If bad cites get caught, the response is immediate disclosure, full apology, and a top-to-bottom re-audit of the brief, not a quiet refile. Clarke’s opinion shows what “quiet refile” costs.

4. Local counsel engagement letters that spell out substantive review obligations. If you serve as local counsel, the engagement letter should say in writing what you will review and bill for that review. If you take local counsel fees, build a real review into your workflow.

5. A client AI-use disclosure at intake. Ask, in writing, whether the client is using AI tools to draft or review pleadings. Document the answer. Set expectations that any client-generated drafts will be treated as raw material only.

The $110,000 number will grab the headlines. The real story is that a court in Oregon just wrote the playbook for how these sanctions escalate when lawyers fumble the response. The next time a judge is deciding whether to go big on sanctions, Clarke’s opinion is on the desk. Assume every court in the country is reading it.