Sony and Warner Sue Anthropic Over Torrented Lyrics

How many times can one AI company be sued over the same training data? For Anthropic, the count keeps rising. Late on Friday, August 29, Sony Music Publishing, Warner Chappell and a group of other music publishers filed suit in the U.S. District Court for the Northern District of California, naming not only the company but CEO Dario Amodei and co-founder Benjamin Mann personally.

The claim

The complaint accuses Anthropic of "a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works on a massive scale" to build its Claude models. Two kinds of material are at issue: the musical works themselves (lyrics and compositions) and books that contain lyrics and sheet music, which the publishers say were obtained by the millions through piracy. They put the number of infringed compositions at "tens of thousands" and ask for "hundreds of thousands of dollars" per work. For scale, BMG's earlier suit against Anthropic covered 493 compositions.

Anthropic was not immediately available to comment.

Is this new, or a rerun?

Both. Concord and Universal Music filed the first lyrics case in October 2023; it is still live and its claims have grown to about $3 billion. In September 2025, Anthropic settled with book authors for $1.5 billion, the largest copyright settlement in U.S. history. The judge in that case drew a line that has shaped everything since: training on lawfully acquired books was fair use, but downloading them from pirate sources was not. In January 2026, a group of music publishers filed a piracy suit built on exactly that distinction. BMG has its own case. Friday's filing is the broadest of the set and the first to name executives individually.

So the question is not whether the legal theory is new. It is whether the music industry can make the 2025 piracy finding pay out a second time, and the structure of music rights makes that easier than it sounds: one song can have a composer, a publisher and a label as separate rights holders, each with standing to sue over the same work.

What is actually at stake

Multiply "hundreds of thousands" per work by "tens of thousands" of works and you get a theoretical ceiling in the billions, above the authors' settlement. The gap between a ceiling and a verdict is always large, but with an IPO planned for the autumn this becomes a contingent liability that has to be disclosed whatever its eventual size.

For businesses that use AI to produce music, advertising or content, the practical point is narrower than the headline: where your provider's training data came from is now a financial risk to that provider, not just an ethical question. When you sign, check whether the plan includes copyright indemnification. Enterprise tiers from the major providers generally do; free and individual tiers generally do not, and that gap is where the exposure sits if a claim ever lands on you rather than on the model maker.

Sources: TechCrunch, Reuters via Yahoo Finance