The Oregon Supreme Court got two AI hallucination problems on the same day and handled both with the patience of an institution that has learned the hard way that fake law is not cute.

On June 4, 2026, Chief Justice Meagan A. Flynn signed two orders involving self-represented litigants whose filings relied on fabricated legal authorities. One order dismissed an original mandamus proceeding. The other struck a response to a petition for review, imposed a $500 sanction, and allowed a corrected filing only with a certification that every source of law cited, quoted, or paraphrased had been verified.

OPB reported that these were the first times Oregon's highest court had addressed AI-generated legal falsehoods directly. That matters because Oregon already had recent lower-court AI citation trouble. The site already has the Bill Ghiorso story from Oregon Court of Appeals sanctions. This is a different incident: the state supreme court confronting fabricated citations from self-represented parties, not a lawyer disciplined by the intermediate appellate court.

One petition got dismissed

In Aldridge v. Tussing, the relators filed a petition for a writ of mandamus in March 2026. The court later discovered that supporting memoranda appeared to cite nonexistent cases and quotations that could not be found in the cited cases. On April 15, the court issued a show-cause order. It directed the relators to confirm each citation under penalty of perjury, explain how the errors got into the filings, and show why the court should not strike the petition or impose other sanctions.

The response made things worse.

The relators acknowledged that several fabricated authorities had appeared in their memoranda and said the errors came from reliance on a generative AI service called LegalAI. That admission might have been the moment to slow down, verify every sentence, and bring the court something sober enough to survive daylight.

Instead, less than 12 hours after receiving the show-cause order, they filed a declaration that included citations to at least four more cases the court said did not exist in the Oregon Appellate Reports or the Oregon Reports. The declaration was filed under penalty of perjury. The court had already warned that sanctions were on the table. The order described this as digging a deeper hole, which is judicial language for "please stop helping yourself."

The court struck the amended mandamus petition and supporting documents, struck the response to the show-cause order, dismissed the proceeding, and dismissed pending motions as moot. The order stated the reason plainly: injecting false precedent undermines the integrity of the proceeding, and doing it repeatedly in response to a show-cause order warrants meaningful sanction.

The other filing got struck and fined

In Witkin v. McGreevy, the problem arose in a response to a petition for review. The petitioner argued that the response relied on authorities that could not be located in the Oregon Reports. The court issued a show-cause order on April 10, 2026. The respondent later addressed the fabricated authorities and explained that generative AI had been used to create the pleading.

The court imposed a $500 financial sanction, struck the response, and gave leave to refile at the respondent's own expense. But the court narrowed that permission. The amended response could not raise new arguments or otherwise be redrafted except to correct inaccurate descriptions of cited authorities. It also had to include a certification that every source of law cited, quoted, or paraphrased in the revised response exists.

That certification requirement is the important procedural lesson. The court did not merely say "try again, but better." It required a verification representation tied directly to every legal source. The implicit message is harsh but fair: if AI helped make the mess, the next filing needs an explicit human cleanup receipt.

Self-represented does not mean consequence-free

Courts usually give self-represented litigants some latitude. People without lawyers often miss procedural niceties. They may use informal phrasing, misunderstand deadlines, or cite authority imperfectly. That leniency has a purpose: access to courts should not depend entirely on being fluent in legal ritual.

But latitude is not immunity. The Oregon Supreme Court drew that line clearly. A person representing themselves still has to make a reasonable effort to avoid affirmative misrepresentations. Submitting fake cases and fake propositions is not a harmless formatting error. It forces the court and opposing parties to spend time proving that the supposed law is vapor.

That time matters. Appellate courts are not customer-support desks for hallucinated citations. Every hour spent chasing nonexistent authorities is an hour not spent resolving real disputes with real law. The court said as much through the Chief Justice's public statement quoted by OPB: when the court must spend substantial time addressing fabricated legal arguments, it comes at the expense of other cases.

Why legal AI is so tempting here

Self-represented litigants are exactly the audience most likely to be seduced by legal chatbots. Legal writing is expensive, intimidating, and full of rituals that look like magic to outsiders. A tool that promises to turn a dispute into polished legal argument feels like access to power.

The problem is that polish is not authority. LegalAI, ChatGPT, Gemini, Claude, and similar systems can make text look lawyerly without making it legally sound. They can invent a plausible Oregon case name, assign it a reporter citation, attach a proposition, and produce a paragraph that looks like it belongs in a brief. If the user does not have the tools or knowledge to verify it, the hallucination enters the court record dressed like precedent.

That is especially dangerous in appellate courts. Appellate arguments depend heavily on cited authority. A trial court can sometimes work around a messy factual submission by asking questions or developing a record. An appellate court is supposed to decide legal questions based on briefs, records, and precedent. Fake precedent poisons the basic workflow.

The answer is verification, not sympathy theater

It is easy to feel some sympathy for self-represented parties who turn to AI because legal help is expensive. That sympathy should not mutate into acceptance of fake legal filings. The harm lands on the opposing party, the court, and everyone waiting for the court's attention.

The useful lesson is not "non-lawyers should never use AI." It is narrower and more practical: nobody should file AI-generated legal assertions without verifying the authorities in an actual legal database or official source. If you cannot verify that a case exists and says what the filing claims, the cite does not belong in the filing. If a chatbot cannot show its work through real sources, it is not doing legal research. It is doing legal cosplay with consequences.

The Oregon orders also show that courts are starting to build procedural responses. Show-cause orders, struck filings, sanctions, dismissal, leave-to-refile restrictions, and certification requirements are all part of the emerging toolkit. Courts do not need a special AI rule for every situation because the old rules already say you cannot lie to the court or file unsupported legal claims. AI just makes the lying easier to do accidentally and at scale.

These two orders are not the largest AI legal sanctions story, but they are a clean warning from a state high court. A fake citation is not less fake because a chatbot made it. A self-represented filing is not safe because the author did not know the tool could hallucinate. And once a court catches the problem, the correct next move is verification, not another machine-polished declaration with fresh imaginary Oregon law inside.