Music publishers, including Sony, EMI, and Warner Chappell, have filed a lawsuit against Anthropic, alleging the company used pirated books to train its Claude AI models. The lawsuit claims Anthropic admitted to pirating over 7 million books, leading to a $1.5 billion settlement, which publishers argue is not enough to deter future infringement. The publishers allege that Anthropic’s illegal torrenting included thousands of books containing lyrics and sheet music to hundreds of copyrighted musical compositions. These actions are said to harm songwriters competing with AI-generated works in music charts, as Anthropic allegedly substitutes artists by reproducing lyrics verbatim and generating new songs that mimic the 'heart' of popular works.
Anthropic’s alleged mass campaign of illegal torrenting began in July 2021, when co-founder Benjamin Mann used BitTorrent to download and upload millions of pirated books from Library Genesis. Dario Amodei, Anthropic’s co-founder and CEO, is also accused of approving the torrenting. Internal messages revealed that Anthropic accessed pirated content through the Pirate Library Mirror (PiLiMi) after LibGen was shut down by the FBI. Mann directed employees to torrent PiLiMi, and an Anthropic staffer responded with, 'zlibrary my beloved,' which publishers point to as evidence of the company’s reliance on piracy. The lawsuit claims that crawling through LibGen and PiLiMi’s catalogs showed Anthropic torrented hundreds of books containing sheet music and song lyrics.
The lawsuit alleges that Anthropic’s training data may have included pirated works, even if the company denies direct use. Publishers argue that AI development often involves a pretraining phase where models use synthetic data or feedback from other AI systems. They claim that Anthropic’s commercial models may have been trained on data derived from pirated books. Additionally, the lawsuit notes that Anthropic used pirated books in its guardrails and that the company destroyed physical books to create unauthorized digital copies of songbooks and sheet music. Publishers allege that Claude models fetch lyrics without sharing copyright information and generate lyrics even when users do not request them. Source: arstechnica
