ChatGPT, in a case against KIPP Texas. An AI-invented quote dropped into a sworn declaration.
Dr. Joy Wilson sued KIPP Texas, Inc. After KIPP moved for summary judgment, Wilson filed a response supported by her own sworn declaration, which tried to show similarities between her former job and the new role at issue.
The declaration quoted language presented as verbatim text from the two job descriptions. KIPP's summary judgment reply pointed out that none of the quoted language Wilson represented as being in her job description actually existed. Wilson's counsel then filed an AI Notice admitting he had used ChatGPT to compare the "Senior Director, Academic Recovery" and "Senior Director, Academic Acceleration" job descriptions, and that the tool "inaccurately presented the content of the job descriptions [...] as language quoted verbatim." In all, counsel had used ChatGPT to generate 11 non-existent statements about the description of Wilson's former job, placed into a declaration sworn under penalty of perjury.
Offered for: an employment-record fact
The court found that counsel violated Local Civil Rule 7.2 by using generative AI without the required disclosure, thereby falsely certifying he had not used it, and violated Federal Rule of Civil Procedure 11 because the artificially generated factual contentions lacked any evidentiary support. Finding counsel solely responsible, the court declined to sanction Wilson herself and sanctioned her counsel: he must reimburse KIPP's attorney fees and costs for preparing its summary judgment reply, and complete 2 hours of CLE on artificial intelligence within four months. The court denied KIPP's motion to strike and granted Wilson leave to file an amended response omitting all AI-generated information.
Kingsfield returned REJECT. The declaration's quoted statement does not exist. The court found no such language in the job description it claimed to quote.
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.