What Sony Music and Warner Chappell allege
The complaint, filed 28 August 2026, alleges Anthropic used 'tens of thousands' of copyrighted musical compositions without a license to train its Claude models. Named works include 'Eye of the Tiger', Marvin Gaye's 'Ain't No Mountain High Enough', and Taylor Swift's 'Paper Rings'. Per the complaint, Anthropic obtained the works through several parallel channels: BitTorrent downloads from the piracy platforms Library Genesis and Pirate Library Mirror, scraping lyrics from licensed providers like MusixMatch and LyricFind, and via large-scale training datasets including Common Crawl, The Pile, and Books3. The plaintiffs are seeking a jury trial and statutory damages of up to $150,000 per infringed work - given the stated scale of tens of thousands of works, a potential total in the billions.
What's actually new: personal liability for the founders
What sets this suit apart from comparable cases against AI vendors is the decision to name CEO Dario Amodei and co-founder Benjamin Mann as individual defendants alongside the company. The complaint states it explicitly: 'Dr. Amodei and Mr. Mann are personally liable for their respective roles in this illegal torrenting.' Specifically, it alleges Mann himself used BitTorrent in June 2021 to download at least five million pirated books from Library Genesis, and that other Anthropic employees downloaded at least two million more works from Pirate Library Mirror in July 2022.
The legal theory behind this targets direct, personal involvement in the underlying conduct, not general responsibility as company leadership. That distinction matters legally: personal liability for founders and executives breaks through the usual limitation of liability to the company itself, making the case markedly more immediate and personal for the defendants, regardless of the financial outcome for the company.
Why the plaintiffs point to an already-settled case
The complaint explicitly builds on an already-concluded precedent: in September 2025, Anthropic agreed to pay $1.5 billion to book authors to settle the class action Bartz v. Anthropic - also over training on pirated works, in that case books rather than music, sourced through the same piracy platforms. At roughly $3,000 per affected book, that settlement was the largest publicly known copyright settlement ever; Anthropic also had to fully delete the affected training datasets. Preceding it was a ruling by Judge William Alsup, who found that training AI on legally acquired books was generally fair use, but explicitly denied a fair-use exception for using pirated works.
Sony Music and Warner Chappell explicitly cite that settlement in their own complaint as an argument: despite the multi-billion-dollar settlement, the company hasn't fundamentally changed its practices, and evidently treats copyright violations as 'just the cost of doing business'. That exact repetition of the same structural conduct - sourcing training data via piracy platforms - with a different category of work is the argumentative basis for pursuing the individuals involved personally this time, rather than limiting the case to renewed corporate liability.
What this means for companies using Anthropic as a vendor
This series has already examined Anthropic from several angles in recent weeks - from its dispute with the US Department of Defense, to the new 30-day retention requirement on its newest models, to how solid its first reported operating profit actually was. This case adds another dimension, this time explicitly legal and financial: it's not a one-off incident, but a recurring pattern in how training data is sourced, one that has already produced a multi-billion-dollar settlement and is now being revisited in a second, potentially even larger case.
For a company using or considering Claude as a strategic AI vendor, this isn't reason for an immediate switch - the outcome of proceedings like this will only become clear over months or years, and a legal dispute against a vendor doesn't automatically affect your own use of the product. But it is a reason to treat your AI vendor's financial and legal risk exposure as its own, recurring evaluation factor, rather than treating a single concluded settlement as a closed matter.
What this means in practice
- When evaluating strategically important AI vendors, treat ongoing and concluded litigation over training-data sourcing as its own, recurring checkpoint - a concluded settlement, as this case shows, is no guarantee that further, structurally similar cases won't follow.
- Track how this specific case develops, particularly whether and how the court handles the founders' personal liability - it could set a precedent signal for similar cases against executives at other AI vendors.
- Review your own AI vendor contracts for what liability and indemnification clauses apply if a model in use is later linked to legally contested training data.
- Don't assume a large, publicly known settlement automatically changes behavior - the reasoning in this new complaint suggests financial settlements alone don't necessarily change data-sourcing practices.
The real value of this analysis isn't predicting the case's outcome - that can't be credibly forecast. It's making visible a pattern that reaches beyond this single case: once an AI vendor has already paid a multi-billion-dollar settlement for a specific kind of training-data sourcing, that's not a closed chapter - it may be the precedent for the next, larger case.