Sony Music Publishing, Warner Chappell, and several other music publishers filed a lawsuit late Friday against Anthropic and co-founders Dario Amodei and Benjamin Mann, alleging the AI lab ran a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works." The complaint, filed in the U.S. District Court for the Northern District of California, accuses Anthropic of "blatant theft" for using thousands of copyrighted songs and sheet music to train its AI model Claude. The case builds on a pattern of intellectual property disputes that have already cost the company heavily.
"We disagree with the publishers' claims and we intend to defend ourselves robustly in court," an Anthropic spokesperson said in an emailed statement.
Piracy allegations and legal history
The publishers claim Anthropic obtained millions of copies of books containing lyrics and sheet music through illegal torrenting. They describe this as "flagrant piracy," arguing that while AI labs and their legal counsel may believe otherwise, acquiring training data through torrenting crosses a clear legal line.
This is not the first IP suit Anthropic has faced. Some of the same lawyers behind this case also represent Concord Music Group and Universal Music Group in a case filed in January. They also led the Bartz v. Anthropic case, in which authors accused the company of using copyrighted works to train Claude. In the Bartz case, a judge ordered Anthropic to pay $1.5 billion after ruling that while it was legal to use copyrighted works for training, it was not legal to obtain that content through piracy.
Differences from earlier cases
The new lawsuit builds on those earlier arguments but is considerably broader in scope. It directly targets the method of acquisition - torrenting - rather than merely the act of training on copyrighted material. For AI for Legal professionals tracking these cases, the distinction matters: the Copyright Act has different bars for unauthorized reproduction versus unlawful distribution, and the publishers here are alleging both.
The complaint also names Amodei and Mann as individual defendants, a departure from the earlier suits. The publishers take the position that founders bear responsibility for decisions made during company formation and the early rollout of Claude's training pipelines.
Why this matters for Legal
Legal professionals in both corporate and litigation practice should watch this case because it tests a question courts haven't fully settled: whether the method of collecting training data can strip away defenses that otherwise protect AI companies. The Bartz ruling established that copyright use for AI training can be legal - but it also created a liability gap for the means of obtaining that data.
For firms advising AI companies on compliance, the practical takeaway is direct. If your client's data acquisition relies on torrenting or other distribution channels - even for material the client believes is fair use - a judge may treat that act as a separate infringement basis. AI for Paralegals documents examining training data provenance, licensing agreements, and "how we scraped it" decisions could not only be exhibit material but the first thing opposing counsel asks for.
Your membership also unlocks:
