The Oregon Court of Appeals published Williams v. Honl on April 22, 2026. Appellant’s attorneys have to pay $8,044.25 to cover the respondent’s fees after filing an opening brief full of AI-generated fabrications. Six of the nine cases cited were either quoted for words that do not exist or used for propositions of law the cases never supported. The court grounded the sanction in ORAP 1.40 and ORCP 17, which together authorize attorney fees as a sanction for filing a brief that falsely certifies its support in existing law.
Of course, this sort of issue is far from new at this point. According to Damien Charlotin’s site, there are now over 1300 cases (about half pro se) in which the lawyer or party used hallucinated cases or case law. I suspect that there are many more cases than that, since Damien admittedly does not know about all cases. The interesting twist in Williams is what the court wrote about how the attorney tried to clean up the mess.
First some background. Appellant’s counsel filed her opening brief. Respondent filed an answering brief pointing out the problems. Counsel then moved to file an amended opening brief. Her motion said the amendment “does not change the assignments of error, issues presented, or relief sought. It corrects the legal standard and citations in light of Respondent’s answering brief.” What appellant left out in her request was the fact that her brief contained fabrications about the law.
In response to Appellant’s request, the court stated that her failure to state that she was correcting hallucinations was a violation of RPC 3.3, meaning a lack of candor. This is not the first time that a court has explained that filing fabricated citations is a lack of candor. It is the first time I have seen a ruling stating that failing to bring attention to the fabrications is in and of itself a violation of 3.3.
Appellant claimed that Appellee was not prejudiced by the filing of the new brief. The court and Appellee both disagreed. The judge agreed with Appellee, noting that Appellee’s lawyers had no choice but to waste time searching for the fabricated cases in the initial brief, and now has to review and respond to the amended brief. The fabrications therefore cost Appellee attorney’s fees that they would otherwise not have had to pay.
The temptation in this situation is obvious. Fix the brief quietly and hope no one asks why. The Oregon court has closed that door. If you file a brief with fabricated authority, the candor rule requires you to tell the court what happened. Not a sanitized description. Not “corrects the legal standard and citations.” You have to say that you used AI, that the cases you cited were fabricated, and that you need to refile. The court will still be unhappy. You will likely face sanctions. But you will not make it worse with a second candor problem. Essentially, the quiet fix buries the evidence of the violation while asking the court to bless the cleanup. In the end, the result was that Appellant had to compensate Appellee for the attorney’s fees related to the extra work.
What to Take from Williams v. Honl
First, the duty of candor is continuing. RPC 3.3 is not a gate you pass through at filing. If you learn after the fact that something you filed is false, you have to correct it, and the correction has to be honest about what you are correcting.
Second, the safe move after you discover AI-generated fabrications in your own brief is immediate and direct disclosure. File a notice. Describe what happened plainly. Ask for leave to refile and accept that sanctions may follow. The Oregon opinion is explicit that the court “appreciate[d] the explicit, if belated, acknowledgment” once the show cause order forced it out. Earlier would have been better.
Third, do not try to rewrite the history in the motion itself. Language like “corrects the legal standard and citations in light of Respondent’s answering brief” reads, after Williams v. Honl, like an attempt to launder the problem. Courts are going to notice.
The filing lawyer is not the only one on the hook. In Oregon, and in many other states, the firm is responsible too. Specifically, ORCP 17 D(1) makes a firm jointly liable for sanctions imposed against its partners, associates, or employees unless joint liability would be unjust. Any firm that permits use of generative AI in research or drafting needs a verification protocol, a disclosure protocol for when something slips through, and training on both. Williams v. Honl is not the last case like this. It is the one that made the candor wrinkle explicit.