US court removes judge whose rulings contained AI-invented content

The last error surfaced at the hearing before the appeals court. Even the "corrected" version of Judge Henry Wingate's order cited a ruling from another federal appeals court, "Cousins v. School Bd. of City of Norfolk." That case does not exist. On October 22, the U.S. Court of Appeals for the 5th Circuit, which covers Mississippi, drew the consequences: it overturned the judge's decision and ordered the case reassigned to another judge.

The case concerns HB 1193, a law Mississippi passed in April 2025 to stop public educational institutions from funding "diversity, equity and inclusion" programs. Teachers, parents, students and associations challenged it on free speech grounds. Judge Wingate, on the bench since 1985, had suspended its enforcement.

Plaintiffs and testimonies that did not exist

The trouble began on July 20, 2025. The judge's first order, which urgently suspended the law, mentioned plaintiffs and defendants who were not part of the case, "statements" from people who do not exist, and false quotations from the text of the law. It also claimed the law had already caused apparently fictitious programs to disappear. When the errors were discovered, the judge replaced the order without explanation. On August 18, 2025, he issued a longer-lasting suspension that reused the corrected version.

Questioned in the fall of 2025 by Senate Judiciary Committee chairman Chuck Grassley, the judge explained that one of his clerks had used the AI tool Perplexity as a "drafting assistant." According to him, the published text was a draft that had not gone through the usual review, and he spoke of a "failure of human oversight."

"Or perhaps, to be charitable, its AI agent"

The opinion, written by Judge Jerry E. Smith, is harsh. Judge Catharina Haynes, one of the other two members of the panel, joins only its final paragraph, which overturns the decision and orders the change of judge. It describes the lower court's handling of the case as "sloppy work, to say the least." It faults the court, "or perhaps, to be charitable, its AI agent," for ignoring the applicable law. To justify the change of judge, the court finds that the embarrassment caused by these hallucinations could push the judge to want to prove he was right anyway. It adds that public trust in the justice system would suffer if the case went back to him.

The plaintiffs' lawyer, Robert McDuff, considers the error serious but notes, according to the local press, that "it's not as if the AI told the judge" to suspend the law. The appeals court did not stop at the AI issue, either: it also found that the lower court erred on the law and clarified the scope of the law, which can now be enforced.

What this case teaches beyond the courts

You do not need to be a judge to fall into this trap. A draft written with AI, then published without review: the same scenario threatens a contract, a legal memo or a response to a call for tenders. A few habits to adopt:

• check every citation, reference or figure produced by AI against its original source, not against the tool's answer;
• clearly separate the draft from the final version in your approval workflows;
• keep a record of AI use in documents that bind the company, so you know what to recheck if an error appears.

European courts are also starting to look into AI errors, like this German judge who barred Google's AI from wrongly accusing a doctor. In Mississippi, the case goes back to the trial court, before a judge who has not yet been named.