Sony Music Publishing, Warner Chappell and a group of other music publishers have sued Anthropic and co-founders Dario Amodei and Benjamin Mann, accusing the AI lab of a “brazen campaign of illegally torrenting, scraping, and downloading copyrighted works.” According to TechCrunch AI, the suit landed late Friday in the U.S. District Court for the Northern District of California and was first reported by Music Business Worldwide. The publishers go further than the usual complaint, calling Anthropic’s use of thousands of copyrighted works to train its Claude model “blatant theft.”
Anthropic isn’t backing down. “We disagree with the publishers’ claims and we intend to defend ourselves robustly in court,” a company spokesperson told TechCrunch in an emailed statement. That sets up another courtroom fight over the single most contested question in AI right now: where training data comes from, and how it was obtained.
What the publishers are claiming
The core accusation isn’t just that Anthropic used copyrighted material. It’s how the company allegedly got it. TechCrunch AI reports the lawsuit accuses Anthropic of “flagrant piracy” through illegal torrenting to pull millions of copies of books, including titles that carry song lyrics and sheet music.
That distinction matters. Courts have started to draw a hard line between two separate acts:
- Using copyrighted works to train a model, which has so far found some legal cover.
- Acquiring those works through piracy, which has not.
The publishers are aiming squarely at the second point.
Why this is bigger than one case
This isn’t Anthropic’s first legal headache, and that’s what makes it serious. Some of the same lawyers behind this filing also represent Concord Music Group and Universal Music Group in a case filed back in January. That legal team also led Bartz v. Anthropic, the authors’ case over copyrighted works used to train products like Claude.
Anthropic was ordered to pay $1.5 billion in that landmark Bartz case. The judge’s reasoning is the piece everyone in the industry should be watching. According to TechCrunch AI, the court ruled it was legal for Anthropic to use copyrighted works, but not legal to acquire that content through piracy. In other words, training on copyrighted material survived. Downloading it from torrent sites did not.
This new lawsuit builds directly on that logic. It’s broader than the earlier cases, and it leans on the same piracy argument that already cost Anthropic more than a billion dollars.
Why it matters for the AI industry
What stands out here is the shift in strategy. Rightsholders spent the early part of this fight arguing that training on their work was itself illegal. That argument keeps running into fair use defenses. So the playbook is changing. Now the pressure point is data provenance, the paper trail of how a lab actually sourced its training corpus.
For anyone building or funding AI models, that changes the risk calculation:
- Provenance is now a legal liability. Where you got the data can matter as much as what you did with it.
- “We just scraped the internet” is a weak position. Torrenting pirated books is exactly the behavior courts are punishing.
- Licensing deals get more attractive. Paying upfront starts to look cheap next to a $1.5 billion judgment and repeat litigation.
Music publishers also bring a specific weapon. Lyrics and sheet music sit inside books and datasets in ways that are easy to trace, which makes infringement simpler to prove than with a general text corpus.
What to expect next
Anthropic has signaled it will fight, so this heads for a drawn out battle rather than a quick settlement, though the Bartz payout shows settlements are on the table when the numbers get large enough. Expect other labs to quietly audit how their own training data was collected, because the same lawyers keep winning on the same argument.
The legal question of using copyrighted works may be softening in AI’s favor. The question of how you obtained them is turning into the industry’s most expensive weak spot. Full details are available at the original TechCrunch AI report.