Key facts
- Sony, EMI, and Warner Chappell are suing Anthropic for alleged copyright infringement.
- The lawsuit claims Anthropic used pirated books, including song lyrics and sheet music, to train its AI models.
- Publishers allege Anthropic's AI models can generate songs that mimic popular works and compete with human artists.
- Anthropic previously settled with book authors for $1.5 billion over similar allegations.
- Anthropic denies the claims, asserting that AI training is transformative fair use.
Music publishers, including Sony, EMI, and Warner Chappell, have filed a lawsuit against AI firm Anthropic, alleging that the company used pirated books containing copyrighted song lyrics and sheet music to train its Claude AI models. The publishers contend that this mass infringement violates their rights and harms songwriters by enabling AI-generated music to directly compete with human artists.
The lawsuit claims that Anthropic engaged in a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works on a massive scale" to develop its AI. Specifically, it points to Anthropic co-founder Benjamin Mann allegedly using BitTorrent to download millions of pirated books from Library Genesis (LibGen) and later directing employees to access a copy of Z-Library through a "Pirate Library Mirror" (PiLiMi). Internal messages, such as one staffer's comment, "zlibrary my beloved," are cited as evidence that Anthropic employees extolled piracy.
Publishers allege that Anthropic's AI models can regurgitate lyrics verbatim and generate new songs that mimic the "heart" of popular works, potentially diluting royalty pools. They also claim Anthropic destroyed physical books to create unauthorized digital copies of songbooks and sheet music collections. While Anthropic denies using pirated works for its commercial models and argues that AI training constitutes fair use, the publishers suggest that even indirect use of pirated data, such as through synthetic data or reinforced feedback from models trained on such data, constitutes infringement.
Anthropic has responded by calling the lawsuit a recycling of allegations from previous cases and asserting that generative AI training is a transformative fair use, citing a court ruling. However, the publishers argue that previous rulings hinged on the inability to prove market harm, which they believe they can demonstrate in the music industry.