What happened

During ongoing litigation in San Jose, California, in which Universal Music Group, Concord and ABKCO accuse Anthropic of using copyrighted song lyrics to train its Claude AI model, a data scientist for Anthropic, Olivia Chen, submitted a court filing that cited a non-existent academic article from The American Statistician journal. 

Apparently created by Anthropic’s chatbot Claude, the citation included fabricated author and title details. The error was discovered when the plaintiffs’ attorney confirmed with the supposed authors and the journal that the article was fictitious, prompting the judge to call the issue “very serious and grave,” and to require Anthropic to formally respond. 

The incident raised concerns about the reliability of AI-generated content in legal proceedings and the potential for misleading the court, which could undermine the integrity of the judicial process. It may also damage Anthropic’s legal credibility. 

Why it happened

According to Anthropic’s legal team, the mistake occurred when their attorney used Claude to format references for a filing, but the AI “hallucinated” the title and authors, resulting in an inaccurate reference. 

The company’s lawyers described the incident as an “embarrassing and unintentional mistake,” emphasising that it was not a deliberate fabrication but a citation error that slipped through manual review.