ChatGPT. A citation assembled from pieces of real but unrelated cases.
Suzanna and Roger Mattox brought a product case against Product Innovations Research and Cosway Company in the Eastern District of Oklahoma. Across eleven pleadings, from summary-judgment briefing to motions in limine and even a sanctions motion, plaintiffs' counsel filed fabricated citations, misquoted authorities, and quotations of law that does not exist.
One example: the briefing cited "Williams v. Borden, Inc., 637 P.2d 731 (Okla. 1981)," which does not exist. A real Tenth Circuit case shares the name but does not contain the quoted law. The court's independent review counted 28 false or misleading citations, 14 fabricated cases and 14 erroneous or misquoted ones. Defense counsel had flagged that several cited cases could not be located.
At a show-cause hearing, signing attorney Harrison Howie admitted he used ChatGPT to make his writing more persuasive, that it changed his citations, and that he filed without verifying them. An earlier motion had described the fake authorities as "clerical and formatting errors."
Offered for: an Oklahoma products-liability rule
The court found Howie and the other counsel of record violated Rule 11(b) through reckless disregard, not intent to deceive. It struck the eleven pleadings and ordered verified amended pleadings. Citing the attorneys' contrition, it imposed reduced fines: $3,000 on Howie, $2,000 on T. Ryan Scott, and $1,000 on Sach Oliver, with public reprimands and a 12-month sponsorship restriction on local counsel Gary Buckles. The Howie and Oliver firms were ordered to jointly reimburse $23,495.90 in defense attorney's fees and costs.
Kingsfield returned REJECT. No case exists at this citation. A similar Tenth Circuit decision exists but contains none of the quoted law. The citation conflates unrelated authorities.
This matter is among the US case-law cases we ran Kingsfield against in the Charlotin benchmark. Kingsfield's existence check confirms every citation against the case-law corpus and an independent source (CourtListener) before any claim resting on it can stand. Across the run Kingsfield ruled 696 of 700 testable claims correctly, including every one of the 504 fabricated citations and 92 fabricated quotes.
Honest scope. This measures Kingsfield's ruling layer given an extracted citation, not an end-to-end run over a full brief. The case record and the AI-tool attribution come from Damien Charlotin's public database and the court order; the verdict is Kingsfield's. See the full 700-claim benchmark.