Sony Music Publishing and Warner Chappell Music have sued Anthropic in California, accusing the Claude developer of unlawfully copying tens of thousands of copyrighted musical compositions. The publishers allege that lyrics and sheet music were taken from pirate libraries and other online sources, then used to train and operate Anthropic’s AI models.
Key takeaways
Sony and Warner’s complaint could become one of the most consequential music and AI copyright disputes yet. The case puts both training data and AI-generated responses under scrutiny.
- The lawsuit was filed on 28 August in the US District Court for the Northern District of California.
- Anthropic, chief executive Dario Amodei and co-founder Benjamin Mann are named as defendants.
- The publishers allege copying involving “thousands if not tens of thousands” of compositions.
- Potential statutory damages could reach $150,000 per work, alongside separate claims concerning copyright information.
- Anthropic says it rejects the allegations and will defend itself in court.
What Sony and Warner allege
The publishers claim Anthropic used torrenting, scraping and downloads to acquire protected lyrics and sheet music. The complaint names unauthorised repositories including Library Genesis and Pirate Library Mirror, while also alleging that material was collected from legal lyric websites and large public datasets.
According to the filing, these works were copied repeatedly, including as training inputs for Claude. The publishers further allege that Claude can reproduce identical or near-identical lyrics when users submit suitable prompts, potentially turning the AI system into a substitute for licensed music content.
The complaint cites songs including Mariah Carey’s “All I Want for Christmas Is You”, Survivor’s “Eye of the Tiger”, Leonard Cohen’s “Hallelujah”, “Uptown Funk”, The Beatles’ “I Am the Walrus” and Taylor Swift’s “Cruel Summer” and “Paper Rings”. The allegations have not been tested at trial.
Billions of dollars could be at stake
Sony and Warner are seeking statutory damages of up to $150,000 for each work found to have been wilfully infringed. They also allege that Anthropic removed or altered copyright-management information, a claim that could attract penalties of up to $25,000 per violation.
The publishers want a jury trial, details of the material used to train Claude and the destruction of allegedly infringing copies. With tens of thousands of compositions potentially involved, the headline exposure could run into billions, although any eventual award would depend on court findings and the number of eligible works.
Anthropic faces a broader copyright battle
Anthropic has rejected the claims, saying it intends to defend itself robustly. The company is expected to argue that training AI models is transformative fair use, pointing to a 2025 ruling that found the use of copyrighted books for training could qualify as fair use in principle.
That ruling also found infringement in Anthropic’s copying and storage of more than seven million pirated books. The company later agreed to a $1.5bn settlement with authors and rights holders, but Sony and Warner argue that the settlement has not deterred further alleged infringement.
The music case follows earlier actions involving Universal Music Publishing Group, Concord Music Group, ABKCO, BMG and Round Hill Music. Together, the disputes reflect a growing confrontation between AI developers seeking vast training libraries and rights holders demanding permission and payment.
Why the case matters for AI users
The outcome could influence whether developers must license lyrics and other musical works before training generative systems. A ruling for the publishers might raise development costs, restrict lyric-related prompts and accelerate commercial licensing agreements.
For readers following AI through Power AI Media’s news, reviews and wider technology coverage, the case is part of a larger question: whether generative AI can scale through unlicensed creative archives, or whether sustainable development will require clearer consent, attribution and compensation rules.