Sony Music Publishing and Warner Chappell Music filed a copyright lawsuit against Anthropic on August 28, 2026, alleging that tens of thousands of copyrighted musical compositions were copied to train the company's Claude AI models. The complaint, filed in the US District Court for the Northern District of California, does not stop at the company: it names CEO Dario Amodei and co-founder Benjamin Mann as individual defendants. Music Business Worldwide characterized the filing as a multi-billion dollar lawsuit, while Unite.AI, which reviewed the complaint, reported that the publishers describe the conduct at issue as "one of the largest and most blatant ongoing thefts of intellectual property in history."
The suit lands as the AI industry's training-data practices face mounting legal scrutiny — a story we continue to track in our latest AI developments coverage.
Four Counts, Two Executives Named Personally
According to the complaint, the publishers — filing collectively as "the Music Publishers" — bring four claims. The first alleges direct copyright infringement through torrenting against all three defendants. The second charges Amodei and Mann personally with contributory infringement for allegedly directing and approving the torrenting activity. The third and fourth counts, aimed at Anthropic alone, allege direct infringement through scraping, downloading, destructive scanning, model training, and AI outputs, as well as the removal or alteration of copyright management information — the song titles, songwriter names, and copyright notices that federal law protects.
The publishers are demanding a jury trial. Songs named in the filing include "Ain't No Mountain High Enough," "All I Want for Christmas Is You," "Eye of the Tiger," "Livin' on a Prayer," "September," "Hallelujah," and Taylor Swift's "Paper Rings."
From LibGen to Songbooks
A substantial portion of the complaint builds on findings from Bartz v. Anthropic, the authors' class action in the same district, where the court found that Anthropic had torrented millions of pirated books and described the conduct as "straightforward piracy but at massive scale." Anthropic settled that case in September 2025 for $1.5 billion.
The new filing alleges that Mann used the BitTorrent protocol in June 2021 to download at least five million pirated books from Library Genesis (LibGen), and that Anthropic employees torrented at least two million more from a successor site called Pirate Library Mirror in July 2022. According to the publishers, those hauls included hundreds of songbooks and sheet-music collections containing their compositions, and that Amodei expressly authorized the downloads. Because BitTorrent users upload pieces of files to others while downloading, the complaint argues each torrent also violated the publishers' exclusive right of distribution.
The complaint quotes internal Anthropic material unsealed in the Bartz case, including Mann's description of LibGen as "sketchy AF," an Anthropic archive team's characterization of the site as a "blatant violation of copyright," and a 2024 planning document about the scanning operation that stated, "We don't want it to be known that we are working on this."
Scraped Lyrics and Destructive Scanning
Beyond torrenting, the publishers allege Anthropic scraped lyrics from licensed lyric websites including MusixMatch and LyricFind in violation of those sites' terms, ran a "destructive scanning" operation that digitized millions of second-hand physical books before destroying them, and drew on third-party datasets including Common Crawl, The Pile, and Books3. The complaint states the publishers have never licensed their works to Anthropic for any of these uses — while noting they have entered licenses permitting authorized uses of their compositions by other AI companies.
Memorization, Guardrails, and "Cleaned" Corpora
The filing describes how the allegedly unlicensed lyrics move through Anthropic's development process. When Anthropic assembles a training corpus, the complaint says, engineers "clean" the text with extraction tools that strip out copyright notices and owner names while leaving the expressive content — a process the publishers characterize as deliberate concealment. The complaint alleges Claude models then memorize lyrics during training and can reproduce them verbatim or near-verbatim in response to user prompts, as well as generate derivatives "in the style of" represented songwriters.
The publishers acknowledge that Anthropic added guardrails intended to stop Claude from reproducing copyrighted text after earlier litigation, but allege those measures are easily circumvented by simply re-prompting the model. The complaint further argues that Claude's ability to write new lyrics creates market substitutes that compete with the publishers' catalogs and dilute the streaming royalty pools from which publishers and songwriters are paid.
What the Publishers Want
The publishers seek statutory damages of up to $150,000 per work infringed where infringement is found willful, and up to $25,000 per violation for removal or alteration of copyright management information. They also ask the court to order Anthropic to destroy all infringing copies of their works under court supervision and to provide an accounting of its training data, training methods, and the specific lyrics and other copyrighted works used to train its models.
The publishers are represented by Oppenheim + Zebrak and Pryor Cashman. "Even the most revolutionary of technologies must develop within the bounds of the law, and Anthropic's Claude models are no different," the complaint reads.
Anthropic's Silence — and What Comes Next
Anthropic had not publicly responded to the suit as of its filing date. The company has previously argued, in earlier litigation over books, that training AI models on copyrighted text can qualify as fair use — a defense that survived in part in the Bartz litigation even as the torrenting claims exposed the company to massive liability.
The new case will test whether that fair-use reasoning extends to song lyrics, one of the most heavily licensed categories of creative text. For the music industry, which has already forced AI music generators into licensing deals and courtrooms in the US and Europe, the suit marks another attempt to establish that catalog-scale licensing — not wholesale copying — is the price of training frontier models on creative works.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
