Anthropic Sued by Sony, Warner Chappell Over Claude Song Piracy

The last major publisher joins the fight

All three major music publishers are now suing Anthropic. On August 28, 2026, Sony Music Publishing and Warner Chappell Music filed suit in the U.S. District Court for the Northern District of California. They accuse the Claude developer of illegally torrenting, scraping and downloading copyrighted songs on a massive scale to train its AI models.18 Three defendants are named: the company, CEO Dario Amodei and co-founder Benjamin Mann.2 The Universal-affiliated publishers already have cases pending, so with Sony and Warner now in court, every major publishing group is litigating against Anthropic.89

The plaintiff list is longer than the headline names suggest. The complaint runs 48 pages, and the plaintiffs are 35 publishing entities tied to the two companies.7 The case is docketed as 5:26-cv-09217.14 Judge Eumi K. Lee has set an initial case management conference for December 16, 2026.5

The works named include some very familiar songs: "Ain't No Mountain High Enough," "All I Want for Christmas Is You," "Eye of the Tiger," Leonard Cohen's "Hallelujah," "Uptown Funk," the Beatles' "I Am the Walrus" and Taylor Swift's "Paper Rings."6 The complaint calls the alleged conduct one of the largest ongoing thefts of intellectual property in history.4

Where the billions come from

The headline figure comes from statutory damages, not a stated demand. The publishers want up to $150,000 for each work infringed willfully. They also want up to $25,000 for each time copyright management information was allegedly removed or altered.3 Reports describe the catalog at issue as "thousands if not tens of thousands" of works, so exposure could reach billions of dollars.6 Business Standard notes that any final award depends on what the court finds.3

The publishers are asking for more than money. They want a jury trial, the destruction of infringing copies and an accounting of Claude's training data.10 The training-data demand may matter most in practice. Anthropic does not publish the full makeup of its training corpus, and a court-ordered disclosure would set a precedent the whole industry would watch.

Four counts, two founders

The complaint contains four counts:

• direct infringement through torrenting, against all three defendants
• contributory infringement, against Amodei and Mann personally
• direct infringement through scraping, training and model outputs, against Anthropic
• removal or alteration of copyright management information, against Anthropic7

The complaint focuses heavily on Mann. As reported by The Verge and summarized elsewhere, it alleges that Mann personally used BitTorrent to download more than five million pirated books from Library Genesis. Other employees allegedly took at least two million more from Pirate Library Mirror.5 One account adds that the Library Genesis downloads were allegedly made at Amodei's direction in 2021.13 The filing also quotes internal material unsealed in earlier litigation, in which Mann described LibGen as "sketchy AF."5

The pirate libraries are only part of the claim. The publishers also allege that Anthropic scraped lyrics from licensed services Musixmatch and LyricFind, used web datasets such as Common Crawl, The Pile and Books3, and scanned physical books.13 They say Claude can reproduce lyrics verbatim or nearly so, and that AI-written lyrics would compete with human songwriters.3

Naming the founders is a pressure tactic, and it has been used before. Anthropic is already resisting it in a parallel case. In the publishers' earlier lawsuit, Amodei filed his own motion arguing that the complaint describes torrenting by others and pleads only contributory conduct against him.17

The Bartz playbook

The publishers' strategy comes straight out of Bartz v. Anthropic. In June 2025, Judge William Alsup ruled that training an AI system on text was fair use, but that downloading and keeping more than seven million pirated books was not.13 Anthropic agreed to a $1.5 billion settlement in September 2025. Judge Araceli Martínez-Olguín gave it final approval in July 2026, and it has been called the largest known settlement in a U.S. copyright case.19

Both sides now cite that ruling. Anthropic's spokesperson said training generative models is "a transformative fair use — as the court held in Bartz."1 The publishers point to the other half of the decision: how the material was obtained. Their complaint is built around the sources and methods Anthropic allegedly used.7 The same lawyers worked on the authors' case.1

The publishers also argue the settlement was too small. The complaint says Anthropic treats $1.5 billion as a cost of doing business, and that the sum is too small to deter a company now aiming for a valuation of roughly $2 trillion in a possible IPO.6 Round Hill's earlier complaint used a different benchmark: a $380 billion valuation from a February 2026 funding round.16 The two numbers reflect different moments and different sources. Both are being used to argue that statutory damages must be large to have any effect.

A cost comparison helps explain why publishers are not taking a books-style deal. Authors in Bartz received roughly $3,000 per work.16 Willful infringement can carry up to $150,000 per work. Even a fraction of that per song, multiplied across tens of thousands of compositions, would be far more expensive for Anthropic than the books settlement.

The "recycling" defense, and where it is accurate

Anthropic says the case recycles allegations already before the courts. A spokesperson called it the third suit from the same lawyers.6 That is accurate as far as it goes. Oppenheim + Zebrak represented Universal, Concord and ABKCO in the first lyrics suit in October 2023, which covered about 500 songs. The firm also brought a second suit in January 2026 covering more than 20,000 works and seeking more than $3 billion.6 BMG sued in March and Round Hill in August, though with different lawyers.6

Music Times reaches a similar view from the other side: the Sony–Warner filing adds no new legal theory, and its importance comes from how many publishers are now involved and how much music they control.8 That reading seems right. The legal questions are already in front of Judge Lee. What changes is the price of losing them. Anthropic's best argument is about procedure. It can push to have the new case handled alongside the earlier ones so that a single set of rulings decides most of the dispute. It has already asked the court to group Round Hill's case with the Concord and BMG lyrics cases.16

The rulings that matter first

The Sony–Warner case is new, but other courtrooms may decide its outcome first. Concord I has reached cross-motions for summary judgment on fair use.11 The publishers argue that lyrics are useful as training data only for generating lyrics, which would make the use non-transformative under the Supreme Court's Warhol decision. They also point to Anthropic's admission that it copied the works in full.17 A ruling against Anthropic on that question would greatly strengthen every later music case.

Anthropic is also pushing back on the outputs theory. In Concord II, it moved to dismiss the output-based infringement claim, noting that a complaint covering about 21,000 works does not identify a single infringing output. It also argued that the copyright-management-information claim protects whole musical compositions, not lyrics used as training text.17 If those arguments win, the Sony–Warner case would effectively become a piracy case. That would shrink it, but not make it go away.

The order in which issues are decided also matters. Judge Lee split discovery in Concord II into two phases. The torrenting claims come first, and the training and output claims wait for the summary judgment ruling in Concord I.11 The publishers also claim that books Anthropic bought, scanned and destroyed included songbooks, and they are fighting to examine those scans in the first phase. Anthropic wants them in the second.11

The track record is mixed for both sides. In March 2025, Judge Lee refused to block Anthropic from training on the publishers' lyrics, finding that money damages could make up for any harm.7 Separately, Anthropic agreed in January 2025 to keep guardrails that stop Claude from producing protected lyrics while that case continued.13

The policy question underneath

The bigger issue is whether "fair use for training" protects how training data was collected. Anthropic's position relies on reading it broadly. The publishers argue that collecting unlicensed files and having a model output protected lyrics are separate violations, outside whatever protection training itself has. One analysis concludes that a quick settlement is unlikely until a judge decides that question.13

Europe may be moving toward the publishers' view. A Munich court ruling from November 2025, which is not yet final, reportedly found that memorized lyrics stored in an AI model count as reproduction, and that the EU's text-and-data-mining exception does not apply to pirated source material.12

The likely outcome is that fair use for training survives as a principle while how data was obtained becomes the main source of legal liability. If so, the music industry's coordinated, catalog-by-catalog cases are designed to make Anthropic's piracy-era data collection too expensive to defend at trial. The goal is to push the company toward licensing, which would mean paying publishers like Sony and Warner for their catalogs. The Mann allegations, the effort to name founders personally and the attack on the size of the Bartz settlement all point the same way.