Sony Music Publishing and Warner Chappell Sue Anthropic Over Alleged Music and Book Piracy

The music industry's slow-burning confrontation with generative artificial intelligence turned into a full-scale legal war on Friday, when two of the world's three largest music publishers sued Anthropic, the company behind the Claude chatbot, over what they call a systematic campaign to train its models on pirated songs and books. The lawsuit, filed in the U.S. District Court for the Northern District of California, joins a growing pile of copyright disputes that have followed every frontier AI company since 2023, but it stands apart in both the commercial weight of its plaintiffs and the detail of its accusations. Sony Music Publishing and Warner Chappell Music, the publishing arms of Sony and Warner Music Group, are not asking a court to referee a gray-area argument about fair use in abstract. They are asking the same court system that produced a landmark $1.5 billion settlement for authors last year to find that Anthropic built a central part of its business on stolen creative works.

Anthropic has repeatedly said its training practices are legal and that it respects creators, and it has already begun signing licensing deals with music companies, including a partnership with Universal Music Group announced in 2024. The new lawsuit is therefore not just a complaint about past conduct. It is a coordinated attempt by the music business to set a price and a legal boundary for how AI companies may use song lyrics, compositions and recordings, and to do so before the courts, rather than through private negotiation, set the terms.

Key Facts

TechCrunch reported on Aug 29 that Sony Music Publishing and Warner Chappell filed the complaint alleging a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works" to train Claude. The Verge reported on the filing the same day, Axios confirmed the details on Aug 29, and Billboard followed with additional reporting on Aug 31. The lawsuit names Anthropic as a company, along with chief executive Dario Amodei and co-founder Benjamin Mann as individual defendants, and covers what the publishers describe as "tens of thousands" of musical compositions, from classic catalog hits to recent releases, whose lyrics and sheet music they claim were ingested without authorization.

The publishers are seeking statutory damages of up to $150,000 for each work they can show was willfully infringed, plus up to $25,000 for each removal of copyright-management information, the metadata that identifies a song's owner and appears in digital files. The complaint leans heavily on Anthropic's own past. It cites the company's $1.5 billion settlement in September 2025 with authors led by science-fiction writer Brian Bartz, which resolved claims that Claude had been trained on pirated books, and it alleges that Mann personally torrented at least 5 million books from unauthorized sources as part of assembling training data. Anthropic's valuation, which investors have marked around $2 trillion in recent fundraising rounds, is cited by the publishers as evidence that the company can afford to pay for the content it uses rather than take it.

Behind the legal language sits a basic financial fact that both sides understand. The market for licensed music publishing rights is measured in the tens of billions of dollars, and streaming platforms already pay publishers for the right to reproduce lyrics and compositions. Sony Music Publishing and Warner Chappell control or administer some of the most valuable catalogs in popular music, and they are explicitly arguing that AI training copies should be treated as reproductions that require a license, not as research use that fair use protects.

Analysis

The bigger picture here is that this lawsuit is the music industry's attempt to avoid repeating the books industry's mistake. When authors sued OpenAI and Anthropic in 2023 and 2024, the cases dragged on for years, produced enormous legal fees, and were ultimately resolved through confidential settlements, of which Anthropic's $1.5 billion Bartz deal was the largest, that effectively turned litigation into a licensing tax without establishing clear precedent. The publishers have watched that play out and are now moving early, together, and with the most valuable assets in their industry, to force a court ruling rather than a settlement. What this really means is that they are betting that the factual allegations here are specific enough to survive a motion to dismiss, because a win at that stage would put enormous pressure on Anthropic to license music the way it already licenses books.

TechCrunch reported on Aug 29 that the lawsuit names chief executive Dario Amodei and co-founder Benjamin Mann as individual defendants.

The individual defendants matter more than casual readers might think. Naming Amodei and Mann is a deliberate escalation that courts take seriously, because it shifts the case from corporate liability toward direct allegations about how training data was assembled. The claim that Mann torrented 5 million books, if it survives scrutiny, would be the kind of fact that moves a judge to allow discovery into Anthropic's data pipelines, and discovery is the phase AI companies fear most, because it opens their closely guarded training methods to inspection. At the same time, the publishers face a real weakness: they must show that Claude's outputs actually reproduce protected expression, not just that the model was exposed to lyrics during training, and courts have so far been skeptical of claims that merely training on copyrighted material, without substantial output copying, violates copyright law.

The strategic timing is also notable. Anthropic is negotiating a massive new computing deal and has been courting enterprise and government customers, many of whom care about copyright risk. A public complaint accusing the company of running a piracy operation, filed on the same weekend its flagship model is being positioned for enterprise deployment, is designed to raise the perceived risk of doing business with Anthropic relative to rivals that have signed licensing deals. In that sense the lawsuit is as much a competitive weapon as a legal claim.

Why It Matters

For songwriters and publishers, this case is a direct test of whether the biggest AI companies will pay for music or take it. If the publishers win, AI training on music will require licenses, creating a new revenue stream for an industry that has seen streaming growth slow and is searching for its next growth line. If they lose, or settle quietly, the practical signal is that frontier labs can absorb music and books into training data and pay only when they are caught. For Anthropic, the stakes are existential in scale, not because $150,000 times tens of thousands of works is a number it cannot contemplate, but because the case threatens the two things investors value most: the defensibility of its data pipeline and the stability of its enterprise pipeline. For the rest of the AI industry, the case is a template. Every major lab has been sued by some creative industry, and the outcome of this one, coming after the books settlements, will shape how the music, film and publishing businesses approach the next generation of models.

Axios confirmed on Aug 29 that the publishers seek up to $150,000 in statutory damages for each willfully infringed work.

Next Up

In the coming weeks, expect Anthropic to respond with a motion to dismiss, arguing that training on copyrighted works is protected fair use and that the publishers have not alleged substantial copying in outputs. The court's ruling on that motion, likely several months away, will determine whether the case proceeds to discovery. Watch also for whether other major publishers, or the recording industry through labels, join the case or file their own complaints, and for whether Anthropic quietly signs a licensing deal with Sony and Warner to make the case go away before it produces rulings. Either path will set a marker for the entire generative AI economy.

The Verge reported on the filing that the complaint alleges Mann personally torrented at least five million books.