A US federal appeals court has vacated a district judge’s injunction against Mississippi’s ban on diversity, equity and inclusion programmes in public education — and ordered the case reassigned away from the judge whose orders were built partly on AI-generated errors, as reported by Reuters on Thursday and confirmed by Bloomberg Law, which found US District Judge Henry Wingate “will no longer oversee that case”. It is believed to be the first time an American judge has been removed from a case because of AI use in his own chambers.
The underlying mistake is older than the ruling. In July 2025, Wingate issued a temporary restraining order blocking the state’s DEI law that — as State Solicitor General Anthony Shults later told the New Orleans-based Fifth Circuit — was “based on names and persons that didn’t exist, and based upon authorities and statements that didn’t exist”. The order named defendants and plaintiffs who were not parties to the case, misquoted the state law, and referenced a nonexistent case, according to the DeSoto County News’ account of Mississippi Today’s reporting. After the state raised the errors, Wingate pulled the order and replaced it with a corrected version. In a letter to Senate Judiciary Chairman Chuck Grassley, he acknowledged that a law clerk had used the AI platform Perplexity “strictly as a foundational drafting assistant”, called the docketed draft “a mistake”, and described the root cause as “a lapse in human oversight” — an early draft with no final review. He has since ordered mandatory independent review of every draft opinion by a second clerk, with cited cases printed from Westlaw and attached to the final order.
The Appellate Court’s Reasoning
The ruling by Judge Jerry Smith’s panel did not turn on the hallucinations alone. The National Law Journal’s report quotes Smith: “Given the embarrassment of the AI hallucinations and mistakes of law in the two TROs and the preliminary injunction, it would be an understandable—though obviously unjust—temptation for the judge to want to find a way to prove that, even with correct law and fact, his initial view should prevail.” The panel also noted a fresh AI-related citation error in the replacement order, and questioned whether it could be confident the use of AI had not affected the lower court’s legal reasoning. The outcome: the injunction is vacated, the case moves to a different district judge, and Mississippi’s law can take effect while the broader litigation plays out — the plaintiffs immediately signalled the fight continues, with lawyer Robert McDuff calling the statute “unconstitutionally vague and potentially quite restrictive of free speech”.
Practical effect aside, the significance is precedent-shaped. Dozens of US lawyers have been sanctioned since generative AI spread for filing fabricated citations — a body of discipline we have tracked as courts have started treating AI-sourced fabrication as a first-class misconduct problem (Connecticut’s prompt-injection sanction was even aimed at a litigant weaponising AI inside filings). But the accountability rarely flows upward into chambers. Here it did: the Fifth Circuit treated an AI-assisted judicial order as grounds to strip a judge of a case, after hearing argument in September where McDuff conceded the errors were “really problematic” while arguing they had not shaped the outcome. Every judge’s order now carries a new vulnerability — if a judge’s citations can be shown to be machine-invented, the order itself becomes contestable. As The Verge notes, the dispute is the latest example of AI derailing court proceedings with made-up citations, bogus quotations and dubious research, and the MIT warning that AI-generated suits could make courts “grind to a halt” now reads less as a forecast than a maintenance schedule (our earlier coverage).
The New Zealand Angle
New Zealand’s courts are ahead of most jurisdictions on paper, and it matters that they are. The judiciary published generative AI guidelines in December 2023 — among the first in the world — with separate rule sets for judges, lawyers and self-represented parties, and the guidance already flags the exact failure mode on display in Mississippi: offshore lawyers filing fictitious AI citations, and New Zealand law librarians reporting rising requests for cases that do not exist. What the Fifth Circuit ruling adds is an enforcement consequence for a judge, not just a lawyer, which is precisely the gap NZ’s guidelines cannot fill by exhortation alone. A Mississippi clerk’s Perplexity shortcut cost a law its moment of suspension and a judge his case; the equivalent lapse in a Wellington or Auckland courtroom would cost a judgment its enforceability. Verification-before-filing is no longer a professional courtesy — it is now, on the Fifth Circuit’s reasoning, part of what makes an order an order.
The deeper shift is that “AI errors” have become a procedural weapon. Mississippi’s state lawyers did not merely appeal the injunction; they used the hallucination record to argue the judge himself should step away. Expect this clause to be quoted in briefs wherever a filing, a ruling or a footnote touched a chatbot: the bar for demonstrating diligence has moved, in both directions, and the first court to lose a case to its clerk’s AI shortcut will not be the last.
Sources: Reuters (headline; full text blocked), Bloomberg Law, National Law Journal, The Verge, Mississippi Today/DeSoto County News, NZ Judiciary
