Skip to content
News TechCrunch Aug 2026

Sony Music and Warner Chappell sue Anthropic over Claude training data

On August 28–29, 2026, Sony Music Publishing and Warner Chappell Music filed a joint lawsuit against Anthropic, naming CEO Dario Amodei and co-founder Benjamin Mann as individual defendants. The suit was filed in the U.S. District Court for the Northern District of California.

What happened

The publishers allege that Anthropic illegally torrented, scraped, and downloaded tens of thousands of copyrighted musical compositions to use as training data for Claude. Works cited in the filing include “Eye of the Tiger,” Marvin Gaye’s “Ain’t No Mountain High Enough,” and Taylor Swift’s “Paper Rings.” Sony and Warner are seeking statutory damages of up to $150,000 per work willfully infringed, plus up to $25,000 for each instance of alleged removal of copyright management information. The total potential liability, depending on how many works are adjudicated, could reach into the billions.

Context

The lawsuit arrives after Anthropic agreed to what was reported as the largest copyright settlement in U.S. history — a $1.5 billion deal with authors and book publishers in September 2025. With all three major music publishers now having taken legal action, the music industry’s approach is coordinated rather than incremental. Universal Music Group had filed earlier; this filing completes the alignment of the major labels against AI companies training on copyrighted audio and lyrics.

The case is distinct from prior AI copyright litigation in one respect: it focuses specifically on the process of data acquisition — the alleged torrenting and scraping — rather than solely on what the trained model outputs. This framing, if it holds in court, would have broader implications for how AI companies structure data pipelines, not just what they do with outputs.

Why it matters for writers

The outcome of this case will shape what training data AI companies can legally use, which in turn affects which capabilities AI writing tools can offer. Models trained on song lyrics and literary prose develop different language capabilities than models trained on web text alone. If publishers successfully establish that licensed data is required for training, the cost structure of building AI writing tools changes significantly.

The case also continues to establish that the creative industries — music, publishing, journalism — are treating AI training data as a licensing issue rather than a fair-use question. Writers and journalists who use AI writing tools are working with products whose underlying data practices are now in active legal dispute.