Music publishers sue Anthropic over how Claude was trained

Sony Music Publishing, Warner Chappell, and a group of other music publishers sued Anthropic and co-founders Dario Amodei and Benjamin Mann in the U.S. District Court for the Northern District of California. The complaint, filed late Friday and first reported by Music Business Worldwide, alleges that the lab ran a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works" to train Claude.

What the complaint alleges

The publishers accuse Anthropic of "blatant theft" by using thousands of copyrighted works in the model's training. Music Business Worldwide, which broke the story, describes the case as a multi-billion-dollar lawsuit and frames it as one of the largest and most obvious ongoing intellectual property thefts on record.

Compared to earlier suits, this complaint is broad. Beyond the music catalog, it accuses Anthropic of "flagrant piracy" through illegal torrenting to obtain millions of copies of books, including works that contain lyrics and sheet music.

The precedent it builds on

This is not the first copyright case against Anthropic. Some of the same lawyers behind this complaint represent Concord Music Group and Universal Music Group in a case filed in January, and they were involved in Bartz v. Anthropic, where a group of authors accused the company of using their copyrighted works to train products like Claude.

Bartz already set the principle this new suit is built on. The court ordered Anthropic to pay $1.5 billion, ruling that using copyrighted works to train a model can be legal, while acquiring that content through piracy is not. The distinction matters: the publishers are not arguing that training on copyrighted works is infringement, full stop. Their case is about how Anthropic got the data.

Anthropic's side

"We disagree with the publishers' claims and we intend to defend ourselves robustly in court," an Anthropic spokesperson wrote in an emailed statement to TechCrunch.

Why it matters

Bartz gave training companies a workable line: what gets you sued is how you acquired the data, not the training itself. This complaint tests that line again, this time with publishers that control a large share of the commercial music catalog, and with claims that stretch from songs to books. If the court follows Bartz, the publishers' remedy depends on proving that the data at issue came from torrents and scraping rather than licensed copies. That is an evidence question as much as a legal one, and it puts data provenance, where a training set came from and how it was obtained, at the center of AI copyright law.