Sony Music and Warner Sue Anthropic Over Alleged Pirated Training Data

Introduction

The copyright fight between generative AI companies and content owners is becoming more specific. Rather than addressing only whether copyrighted material can be used to train a model, a new lawsuit against Anthropic focuses on the route by which that material was allegedly obtained.

Sony Music Publishing, Warner Chappell, and a number of other music publishers filed the case in the U.S. District Court for the Northern District of California. The defendants include Anthropic and co-founders Dario Amodei and Benjamin Mann. The publishers describe the alleged conduct as a “brazen campaign” involving the torrenting, scraping, and downloading of copyrighted works.

According to the allegations, Anthropic used thousands of copyrighted works to train Claude. The company has rejected the publishers’ claims and said in a statement that it intends to defend itself robustly in court.

Key points

• A broad group of plaintiffs. The action brings together Sony Music Publishing, Warner Chappell, and other music publishers, reflecting a wider concern about the use of music-related content in AI development.
• Acquisition is central to the case. The publishers are not merely challenging model training. They specifically allege that Anthropic obtained source material through illegal torrenting, scraping, and downloading.
• Lyrics and sheet music are part of the dispute. The complaint reportedly points to millions of copies of books, including works containing lyrics and sheet music, potentially widening the discussion beyond conventional music files.
• The case follows earlier litigation. Some lawyers involved also represent Concord Music Group and Universal Music Group in related proceedings, and worked on the authors’ Bartz v. Anthropic case.
• A prior ruling drew a legal distinction. In the Bartz case, a judge ruled that Anthropic’s use of copyrighted works for training was legal under the circumstances at issue, while acquiring those works through piracy was not. Anthropic was ordered to pay $1.5 billion.

Why it matters

The lawsuit highlights a distinction that may become central to AI copyright compliance: a potentially lawful use of training material does not necessarily make the method used to obtain that material lawful. For model developers, provenance records, download channels, internal copies, licenses, and audit trails could become as important as the architecture or performance of the model itself.

For music publishers, lyrics and sheet music are commercially valuable copyrighted assets rather than incidental text. A ruling accepting the publishers’ allegations could encourage more scrutiny of AI data supply chains, licensing arrangements, and content controls. A ruling favoring Anthropic, meanwhile, could help clarify how courts distinguish training, copying, caching, and the purchase or acquisition of data.

The allegations have not been finally decided. Still, the case shows how AI copyright disputes are evolving. The central question is no longer limited to whether a model may learn from protected works. It increasingly includes where the works came from, how they were obtained, and whether the developer can demonstrate a compliant chain of custody.

Source: TechCrunch AI