Photo Credit: Claude / Anthropic
Major publishers push for summary judgment against AI giant Anthropic for its use of copyrighted song lyrics to train its Claude model.
Universal Music Group (UMG), alongside fellow music publishers Concord and ABKCO, has asked a California judge to rule that U.S. copyright law does not shield AI company Anthropic from liability for having used their song lyrics without permission or compensation to train its Claude AI chatbot model.
Filed on Monday in San Jose federal court, the publishers push back against tech companies’ continued claim that “fair use” applies to copying millions of copyrighted works without authorization to train AI models. Their case argues that Claude’s AI-generated lyrics are by definition derivatives of the publishers’ lyrics and “compete with and dilute the market” for them.
The publishers also argue that Anthropic systematically scraped and copied millions of lyrics and compositions without permission or compensation to build a commercial product that competes with the original works.
Crucially, they note that this is not a “referendum on AI technology,” as they have and do license works to other AI companies—but the endless unauthorized reproduction and generation of “limitless AI rip-offs” sits at the heart of the matter.
“The evidence in this case is overwhelming,” the filing asserts, adding that Anthropic has “committed copyright infringement on a massive scale.”
The publishers first sued Anthropic in 2023 with allegations that it infringed their copyrights in lyrics from at least 500 songs by major artists like the Beach Boys, the Rolling Stones, and Beyoncé.
It’s just the latest filing among dozens of cases by copyright holders and authors against tech giants and AI startups, including OpenAI, Meta, Microsoft, and Google. Anthropic became the first major AI company to settle such a case last year, agreeing to pay a group of authors $1.5 billion to resolve their class action.
The majority of these cases center on whether AI systems can claim fair use of copyrighted material by using it to create “transformative content.” But like the music publishers, other copyright holders have argued this cannot be the case, given that the output of models like Claude are inherently derivative of the content on which they’re trained, and do not produce “new” content from whole cloth.
Interestingly, while Anthropic has denied the allegations lobbed against it, the company has not yet argued for fair use in the music publishers’ case. However, its case settled with the authors last year hinged on a San Francisco judge’s ruling that the use of books for AI training was “quintessentially transformative.”
