Sony Music Publishing and Warner Chappell have filed an Anthropic lyrics lawsuit against Anthropic in a Northern California federal court, accusing the company of building Claude‘s training data from pirated song catalogs. The complaint names chief executive Dario Amodei and co-founder Benjamin Mann personally, alongside a “brazen campaign of illegally torrenting, scraping, and downloading copyrighted works on a massive scale.”
The publishers list specific songs in the filing, including “Eye of the Tiger,” “Hallelujah,” “September,” “Livin’ On a Prayer” and “Great Balls of Fire,” along with compositions by Mariah Carey and Taylor Swift. Sony and Warner Chappell point to Library Genesis and Pirate Library Mirror as the source archives, the same shadow libraries at the center of the $1.5 billion settlement Anthropic reached with a group of authors over book piracy. The publishers are seeking a jury trial and statutory damages of up to $150,000 per composition for willful infringement.
The Munich precedent behind the Anthropic lyrics lawsuit
A Munich court ruled against OpenAI in November 2025 in a comparable case, finding that memorizing lyrics inside a model constitutes reproduction and that outputs reciting those lyrics amount to communication to the public. The court held that the EU’s text and data mining exception did not cover the practice, since permanent memorization goes beyond transient analysis and the rightsholder had opted out. European copyright law also requires lawful access to the source material, and pirate libraries do not meet that standard.
That ruling gives Sony and Warner Chappell a legal template outside the United States, even though this new case runs through American courts under different statutes. Both disputes turn on the same underlying question: whether training a model on copyrighted lyrics, regardless of how the outputs are used, counts as infringement on its own.
What the EU AI Act requires that US law doesn’t
The EU’s AI Act requires providers of general purpose models to maintain a copyright policy and publish summaries of their training data, with enforcement handled in Brussels. That creates an asymmetry between the two markets. American rightsholders, including Sony and Warner Chappell, have to sue first and rely on discovery to learn what material a company like Anthropic actually used. European rightsholders are entitled to that information without filing a complaint.
Anthropic has not filed a public response to the new complaint. The case adds to a growing list of copyright disputes facing AI developers. OpenAI, Meta, and Google face similar claims from publishers and authors over how their models were trained, and courts on both sides of the Atlantic are still working out where the line between research and infringement sits.
For Anthropic specifically, the timing matters. The company only recently closed its book-piracy settlement with authors, and a second high-value case covering music lyrics puts fresh scrutiny on how it sourced training data before that agreement. Statutory damages of $150,000 per composition, multiplied across the dozens of songs named in the complaint, could put the total exposure well into the hundreds of millions of dollars if the publishers prevail at trial.