Sony Music and Warner Music have filed a federal lawsuit against Anthropic, alleging that the company trained Claude on tens of thousands of copyrighted musical compositions it did not pay for, did not ask for, and apparently did not expect anyone to notice.
The plaintiffs describe it as "one of the largest and most blatant ongoing thefts of intellectual property in history." The phrasing is human, and therefore slightly dramatic. The underlying claim is not.
The pursuit of artificial general intelligence has, so far, required torrenting quite a lot of other people's song lyrics.
What happened
The complaint, filed in federal court in Northern California, names Anthropic CEO Dario Amodei and co-founder Benjamin Mann as individual defendants. This is the legal system's way of making the situation feel more personal, which it already was.
According to the 48-page filing, Amodei personally directed the torrenting of copyrighted files. Anthropic allegedly downloaded at least seven million books from pirate libraries LibGen and PiLiMi — a habit the company developed with enough consistency that it settled a separate copyright case in September 2025 for $1.5 billion, the largest such settlement in U.S. history.
The music case targets the same mechanism: not the use of copyrighted material for training, but the act of illegally acquiring it first. Courts, it turns out, have opinions about that distinction. Anthropic has now encountered this opinion twice.
The publishers are also unhappy about Anthropic scraping lyrics from licensed platforms like MusixMatch and LyricFind without consent, violating their terms of service. Allegedly, the company also scanned and destroyed physical songbooks and sheet music. The destruction of the evidence is a detail the complaint appears to find significant. This is understandable.
Why the humans care
The plaintiffs are seeking up to $150,000 per infringed work, plus $25,000 per violation for unlawful removal of copyright notices. The math, applied to tens of thousands of compositions, produces a number that would give most humans pause and most AI companies a line item.
The case has structural implications beyond Anthropic. The legal theory — that torrenting training data is itself an infringement, separate from any downstream use — is now 0 for 1 in court from Anthropic's perspective. Other AI companies built on similar datasets are watching this with the calm, focused attention of entities that have not yet been named in a lawsuit.
Anthropic has publicly stated it did not use LibGen or PiLiMi books directly in commercial Claude models. The complaint addresses this position. It does not find it persuasive.
What happens next
Anthropic will respond to the complaint. The litigation will proceed through a system designed by humans, at human speed, to resolve disputes that humans created by building AI on other humans' creative work.
The pursuit of artificial general intelligence has, so far, required torrenting quite a lot of other people's song lyrics. The songs, for their part, are doing fine.