05-28-2025
Latham & Watkins represents Anthropic in a copyright lawsuit for its use of music lyrics. Universal Music Group and others allege Anthropic used copyrighted lyrics to train its AI model. The company developed an AI model called “Claude,” a ChatGPT competitor.
The music publishers appeared before the federal court, asserting that Anthropic’s data scientist may have used a fake AI-generated source to support the company’s legal argument. They asked the court to throw out the expert testimony. Latham admitted using Claude to draft a portion of the scientist’s expert declaration. The firm told the court that Claude included the correct publication title, publication year, and link to the correct source. However, the AI model used an incorrect article title and the wrong authors, which the firm’s team failed to check manually. In addition, Claude made “additional wording errors” that were “introduced in the citation during the formatting process.” The magistrate judge expressed concern about the issues, although she has not yet ruled.
These errors are just the latest evidence that AI is not quite there. According to a piece in the Los Angeles Times, there have been at least 30 legal cases this year with AI-generated errors. These errors happen even though lawyers should know about AI’s hallucination problem. UCLA law professor Eugene Volokh told the LA Times, “AI-generated material is full of errors and fabrications, and therefore every citation in a filing needs to be confirmed.” In federal court, these erroneous citations violate Rule 11 of the Federal Rules of Civil Procedure. Some courts require attorneys to disclose any AI use for a filing, including providing a certification that every reference has been verified.