ChatGPT cited a non-existent court judgment in a US federal-court filing; counsel reproduced the failure and the citation does not exist.
ChatGPT cites non-existent court judgment in federal filing
Counsel filing an opposition brief in the Southern District of New York included three case citations generated by ChatGPT. Two of the three citations do not appear in any public legal database. The hallucinated citations include a plausible-looking docket number and a paraphrased ratio that maps to no actual judgment.
The failure was reproduced across three separate prompts asking for "case law on the duty of care for AI-generated content". Each run produced a different non-existent citation alongside one real case.
Cross-model verification: GPT-4o reproduced the hallucination twice; Claude 3.5 Sonnet declined to provide citations without a verified source; Gemini 2.5 Pro declined.
The provider was contacted with a 24-hour right-of-reply window. Response received and appended below.
Why this is an AI incident
A counsel relying on a publicly-accessible legal database alone would have surfaced the missing citation immediately. The hallucination originated in the AI's output, was then incorporated into the filing, and would not have occurred had the AI not been used as a citation generator.
Counterfactual "but-for" test per the Editor's Guide.