Mississippi Federal Judge Uses AI to Write Ruling with False Precedent

On July 20, 2025, Henry T. Wingate, a federal district judge for the Southern District of Mississippi, issued a temporary restraining order halting a state law that restricted diversity programs in public schools. The state attorney general pointed out that the ruling misidentified the original plaintiff and defendant, misquoted state law, and cited non-existent precedents. The judge subsequently removed the original text from the public record and replaced it with a corrected version, initially claiming it was a typographical error.

After Senate Judiciary Committee Chairman Chuck Grassley sent an inquiry, Wingate acknowledged in October that a judicial assistant had used Perplexity as a drafting assistant, and the filed document was an unverified draft. In August 2026, the Fifth Circuit Court of Appeals requested both parties to prepare arguments on whether the case should be reassigned to other district judges due to the use of AI. New Jersey federal judge Julien Xavier Neals rejected a motion to dismiss in the CorMedix securities group lawsuit, with his opinion containing fabricated citations, non-existent precedents from the Southern District of New York, and misstatements of other case outcomes; this opinion had been cited as supplementary authority in another group lawsuit. Neals later admitted that a law school intern had unauthorizedly used ChatGPT for research, and the opinion was withdrawn after being filed due to human error.

In 2023, a New York lawyer was fined $5,000 for fabricating precedents using ChatGPT. A study conducted by Northwestern University and the New York City Bar Association in December 2025 surveyed 502 out of 1,738 federal judges, with 112 responding: over 60% had used at least one AI tool, only 22.4% used it weekly or daily, and 38.4% had never used the listed tools in their work; the primary use was for legal research. In a custody dispute over a Labrador dog named Kyra in a family court in San Diego, California, one lawyer cited a non-existent pet precedent, and the opposing lawyer included the fictitious case in a proposed order, which the judge signed. The appellate court ruled that reliance on fabricated precedents constituted an abuse of discretion, but since the losing party drafted and submitted the order themselves, it was considered a waiver, and the original ruling was upheld, with an additional $5,000 fine imposed on the lawyer who insisted on the false citation, along with a report to the state bar association.

In market mechanisms, the buyers are the judges' chambers with backlog cases and parties who cannot afford lawyers, while the sellers are general models like Perplexity and ChatGPT, as well as legal-specific tools like Westlaw. The incident was driven by the erroneous ruling being filed, with sanctions and reassignment pressures falling on unverified judges' chambers and lawyers, while platforms continue to charge based on search volume. The beneficiaries are legal databases that can enforce verification as a mandatory process, while the pressured parties are the networks of precedents in common law systems that may be cited in subsequent litigation once a fictitious case enters the record.

Source: Public Information