Sony and Warner Sue Anthropic Over Claude Training Data

Sony Music, Warner Music, and other major publishers filed the lawsuit in a Northern California federal court. Uniquely, the complaint names Anthropic CEO Dario Amodei and co-founder Benjamin Mann as individual defendants, accusing them of personally directing the torrenting of copyrighted files. The publishers are seeking damages of up to $150,000 per infringed work, alongside up to $25,000 per violation for the unlawful removal of copyright management information.

The lawsuit focuses on how Anthropic obtained its training data, alleging the company scraped lyrics from licensed platforms like MusixMatch and LyricFind, and utilized controversial datasets such as Books3, The Pile, and Common Crawl. The plaintiffs also claim Anthropic scanned and destroyed physical songbooks and sheet music. Furthermore, the complaint targets Anthropic's use of pirate libraries LibGen and PiLiMi. Although Anthropic denied using these libraries for commercial Claude models, the publishers argue the company trained commercial models on synthetic data generated by non-commercial models that had ingested the pirated texts.

This legal battle follows Anthropic's historic $1.5 billion settlement in September 2025 over the use of pirated books. For AI practitioners and developers, this case underscores a massive shift in legal risk. Courts are increasingly focusing on the acquisition phase of data gathering rather than just the final model output. A recent November 2025 ruling by the Munich Regional Court already established that storing copyrighted lyrics within a model's parameters constitutes reproduction, holding developers liable for outputs even when prompted by users to generate infringing material. As a result, AI teams must implement rigorous, verifiable data-provenance pipelines to avoid catastrophic liabilities and personal executive lawsuits.