Sony and Warner sue Anthropic's founders over alleged training-data piracy

Sony Music Publishing and Warner Chappell accuse Anthropic and its founders of copying tens of thousands of compositions for Claude and seek up to $150,000 per infringed work.

By · Published

Primary source: Axios

Why it matters

The complaint names two Anthropic founders in direct and contributory infringement claims, raising the prospect of personal liability and discovery into executive decisions about how an AI lab acquired training data.

Stylized silhouettes of a vintage record player and sheet music facing a judge's gavel, under a bright sunburst.

Axios reported that Sony Music Publishing and Warner Chappell Music sued Anthropic and co-founders Dario Amodei and Benjamin Mann on August 28, alleging that Anthropic used torrenting, scraping, and downloaded copyrighted works to develop and operate Claude. The allegations have not been adjudicated.

The lawsuit was filed in the U.S. District Court for the Northern District of California. In the 48-page complaint, the publishers allege that Anthropic and its founders acquired copyrighted books, lyrics, and musical compositions without authorization, then used the material to train and operate Claude. The publishers say the case involves tens of thousands of compositions and seek up to $150,000 per work for willful infringement.

The founders are unusually central to the case. Amodei, a Princeton-trained biophysicist and former Google Brain researcher, led large-language-model research at OpenAI before starting Anthropic. Mann spent six years as a Google software engineer and later worked at OpenAI and the Machine Intelligence Research Institute. Anthropic, a San Francisco-based public benefit corporation, emerged from a group of former OpenAI employees in 2021 with a pitch built around safer and more controllable AI systems.

The plaintiffs allege that Anthropic and its founders illegally acquired copyrighted works to develop, operate, and profit from Claude. More specifically, Sony and Warner's publishing units accuse Amodei of authorizing the acquisition of pirated material and Mann of personally downloading some of it, then directing Anthropic employees to collect more.

Anthropic faces scrutiny over how it acquired training data

The complaint alleges that Mann used BitTorrent in June 2021 to download at least 5 million books from Library Genesis, known as LibGen. It says Anthropic employees downloaded at least another 2 million books from Pirate Library Mirror in July 2022. The publishers allege that those collections included lyrics, sheet music, and other copyrighted material they control.

Those totals draw on the record in the separate book-authors case Bartz v. Anthropic, where court records described an Anthropic co-founder downloading 196,640 books from Books3 and later acquiring millions more through LibGen and Pirate Library Mirror.

Sony and Warner also accuse Anthropic of scraping lyrics from authorized display services including MusixMatch and LyricFind, scanning second-hand books, and collecting material from Common Crawl, The Pile, and Books3. These remain allegations by the plaintiffs.

The publishers say the case involves tens of thousands of compositions. The complaint names songs including "Ain't No Mountain High Enough," "All I Want for Christmas Is You," "Eye of the Tiger," "Hallelujah," "Livin' on a Prayer," "September," and Taylor Swift's "Paper Rings." The plaintiffs seek up to $150,000 per infringed work for willful infringement and up to $25,000 for each violation involving the removal or alteration of copyright-management information.

The complaint brings four counts: direct infringement through torrenting against all three defendants; contributory infringement through torrenting against Amodei and Mann; direct infringement against Anthropic covering other alleged copying; and removal or alteration of copyright-management information against Anthropic.

Naming Amodei and Mann gives the plaintiffs a way to seek evidence about the founders' alleged roles in acquiring and using the material. The case could put discovery pressure on internal records showing where training material came from, who approved its acquisition, and what licenses covered its use.

Claude's outputs are also part of the complaint

Sony and Warner allege that Claude can reproduce protected lyrics verbatim or nearly verbatim, including after repeated prompts. The complaint's exhibits identify compositions the publishers say are implicated by Anthropic's training and output practices.

In their demand for relief, Sony and Warner request an accounting of Claude's training data and methods and the destruction of infringing copies under court supervision.

Anthropic has already litigated the acquisition question

A federal judge in Anthropic's earlier book case drew a distinction between training on lawfully acquired books and retaining pirated copies. In Bartz v. Anthropic, the judge ruled that using lawfully acquired books to train language models could qualify as fair use, while downloading and retaining pirated copies in a central library was not excused by fair use.

Anthropic agreed in September 2025 to pay at least $1.5 billion to settle claims from authors and publishers. The Authors Guild's settlement FAQ says the agreement covered roughly 500,000 works and required Anthropic to delete books acquired from pirate libraries.

The music publishers are building on the acquisition side of the earlier ruling. Their complaint alleges several acts: obtaining files through torrenting and unauthorized scraping, retaining those files, copying them during training, and reproducing protected lyrics through Claude outputs. The publishers will need evidence connecting particular compositions and defendants to each alleged act.

Anthropic is far larger now than it was when the alleged downloads occurred. In its May 2026 financing announcement, Anthropic said it raised $65 billion at a $965 billion post-money valuation and had crossed $47 billion in run-rate revenue. Those are company-reported figures rather than audited results.

The publishers still must match each claimed composition to a valid copyright and a particular alleged act of infringement. They must also prove the roles they assign to Amodei and Mann. That work-by-work and defendant-by-defendant accounting will determine whether the complaint's large numbers translate into liability for the company and its founders.

Reader comments

Conversation for this story loads after sign-in.