Anthropic already settled a $1.5 billion lawsuit over pirated books. Now, two of the most powerful music publishers in the world want their own reckoning. Sony Music Publishing and Warner Chappell have filed a sweeping federal lawsuit against Anthropic, claiming the company illegally scraped tens of thousands of copyrighted song lyrics to train its Claude models. The damages ceiling: $150,000 per infringed composition, plus $25,000 for each instance of stripped metadata. With tens of thousands of tracks named, the total exposure could reach into the billions.
The 48-page complaint was filed in the U.S. District Court for the Northern District of California. It names not just Anthropic as a corporate entity but also CEO Dario Amodei and co-founder Benjamin Mann personally. The songs at issue span decades and genres, from Mariah Carey’s “All I Want for Christmas Is You” and Survivor’s “Eye of the Tiger” to tracks by Taylor Swift, The Beatles, and Bon Jovi. This is not a narrow, technical filing. It’s a wide-angle attack.
The core allegations are serious. Sony and Warner claim Anthropic didn’t just passively ingest lyrics from the open web. According to the complaint, the company actively acquired copyrighted text through unauthorized torrents, pirate databases including Library Genesis, and aggregators like Common Crawl. The publishers also allege that Claude was allowed to reproduce verbatim lyrics in response to user prompts, which they frame as direct competition with the original works. And they claim Anthropic stripped metadata during processing, removing attribution from songwriters in the process.
Anthropic’s position is predictable but not without legal weight. A company spokesperson called the suit a recycling of existing claims and argued that training generative AI on third-party text qualifies as transformative fair use. That defense has held up in some contexts, but courts are increasingly skeptical when the AI output closely mimics or reproduces the original. Verbatim lyric reproduction is a much harder case to win than general stylistic influence.
The broader context here matters. This lawsuit comes after Anthropic’s $1.5 billion settlement with authors over pirated books, a number that sounds large but is modest relative to the company’s reported $2 trillion valuation. Sony and Warner explicitly cited that settlement in their filing, arguing that a company at that scale treats copyright payouts as a cost of doing business rather than a real deterrent. That framing is designed to push courts toward maximum damages, not negotiated settlements.
For the AI industry, this case is a stress test. Anthropic is not alone in having trained on data scraped from the web. OpenAI, Google, and Meta face similar questions about their training pipelines. But the specificity of this complaint, named tracks, named executives, named pirate sources, makes it harder to dismiss as abstract legal theory. If Sony and Warner win, or even extract a large settlement, every AI lab with music-adjacent training data will need to reassess its exposure fast.
