Sony Music Publishing, Warner Chappell Music and other music publishers have sued Anthropic and two of its co-founders, alleging that the AI developer unlawfully obtained copyrighted songs and other materials to build its Claude models.
The complaint, filed in the U.S. District Court for the Northern District of California, names Anthropic CEO Dario Amodei and co-founder Benjamin Mann as defendants alongside the company. The publishers allege that Anthropic used torrent networks, online archives and scraped websites to assemble training data containing protected lyrics and musical compositions without permission.
The allegations concern both the acquisition of training material and Claude’s behavior after training. The publishers claim Anthropic downloaded works from sources including Library Genesis and Pirate Library Mirror, which have been associated with unauthorized copies of books. They also allege that lyrics were collected from services such as MusixMatch and LyricFind, which license music content.
Although some of the disputed files were books, the complaint argues that those books contained protected lyrics and sheet music controlled by the plaintiffs. The works identified in the case include Marvin Gaye and Tammi Terrell’s “Ain’t No Mountain High Enough,” Bon Jovi’s “Livin’ on a Prayer,” Earth, Wind & Fire’s “September,” Leonard Cohen’s “Hallelujah” and Taylor Swift’s “Paper Rings.”
The publishers further contend that Claude can return identical or closely matching passages from copyrighted songs when prompted. That claim raises a distinct issue from whether Anthropic was entitled to use the works during model development: even if some training uses are found to be fair use, an AI system’s reproduction of protected expression may still create separate infringement risks.
The plaintiffs are seeking statutory damages and a jury trial. U.S. copyright law permits damages of as much as $150,000 for each work when infringement is found to be willful. The complaint also seeks damages tied to the alleged removal of copyright-management information. Because the publishers say tens of thousands of compositions are implicated, the theoretical maximum exposure could reach billions of dollars, although any eventual award would depend on what claims and works survive litigation and how a court assesses liability.
Anthropic did not immediately provide a public response to the complaint.
The case follows earlier litigation examining the difference between training an AI model on copyrighted material and obtaining that material from unauthorized sources. In a lawsuit brought by authors, a federal judge concluded that certain uses of books for AI training could qualify as fair use while treating the retention of pirated copies as a separate copyright problem. Anthropic later agreed to a $1.5 billion settlement in that dispute.
That distinction is likely to be central to the music publishers’ case. Generative-AI developers typically require very large collections of text and other media to train general-purpose models. Courts are now being asked to decide not only whether computational analysis can be transformative, but also whether the origin and storage of the underlying copies independently violate copyright law.
Music rights add another layer of complexity because a song can involve separate copyrights in its composition, lyrics and sound recording, with different entities controlling each right. This complaint focuses on publishing rights covering compositions and lyrics rather than solely on ownership of recorded performances.
Other music companies have also pursued Anthropic over alleged use of protected songs, making the new filing part of a broader effort by rights holders to scrutinize how Claude’s datasets were assembled. The Northern District of California case will now move through the early stages of litigation, including Anthropic’s formal response and potential challenges to the publishers’ claims.
Sources: Anthropic, AI copyright lawsuit, AI copyright lawsuit