ChatGPT, in a brief naming the Attorney General. Two dozen invented quotes in a single filing.
Heriberto Perez-Castillo, a Mexican citizen facing removal, petitioned the Seventh Circuit to review a Board of Immigration Appeals order denying cancellation of removal. His opening brief was riddled with AI hallucinations.
Reviewing the case, the court found that roughly half the cited cases either did not exist or were falsely labeled as Seventh Circuit precedent, and nearly every quotation could not be traced to a real opinion. From pages 19 to 29, every quote appeared to be an AI hallucination. All told there were twenty-four unattributable or outright false quotes and at least seven mislabeled or nonexistent cases, plus factual claims contradicted by the record, including a claim about "U.S. citizen children" still in school when all were over twenty-five.
Counsel Abdullah Salah blamed contract attorney Farah Chalisa, who said she ran the brief through ChatGPT for grammar and stylistic review. After the court ordered an explanation, Salah filed a revised brief the night before argument that swapped in real citations but kept unsupported assertions and at least two hallucinated quotes.
Offered for: immigration-procedure holdings
The court denied the petition for review on the merits. It imposed a $5,000 fine on attorney Abdullah Salah under Federal Rule of Appellate Procedure 46(c) for signing and submitting two briefs without checking the AI-hallucinated citations, legal propositions, and factual representations, and warned that a future incident could lead to disbarment before the court. It declined to sanction Farah Chalisa at this time but admonished her, and referred the matter to the Attorney Registration and Disciplinary Commission of Illinois, directing that the show-cause order remain open until the ARDC reviews it.
Kingsfield returned REJECT. About 24 fabricated or untraceable quotations in one opening brief. At least seven cases were mislabeled as supporting points they do not.
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.