Music publishers are escalating the fight over how AI companies obtain and use copyrighted material. In a lawsuit filed Friday, Sony, EMI, and Warner Chappell alleged that Anthropic’s earlier book piracy issues also involved copyrighted music material, including books containing lyrics and sheet music.
The complaint links the dispute to Anthropic’s historic settlement with authors, in which the Claude maker paid authors $1.5 billion after admitting to pirating more than 7 million books to train AI. The publishers argue that amount was not enough to discourage similar conduct, especially given what they described as Anthropic’s $2-trillion-dollar valuation.
Why music publishers are targeting Claude
The publishers allege that Anthropic’s illegal torrenting included “thousands upon thousands” of books containing lyrics and sheet music to hundreds or more copyrighted musical compositions. Their concern is not limited to possession of those materials. They argue that the works may have been used in ways that affect songwriters and music publishers directly.
According to the complaint, pirated books that Anthropic allegedly plans to keep “forever” include titles featuring the complete works of the Beatles, Taylor Swift’s “best” songs, and “VH1’s 100 Greatest Songs of Rock & Roll.” The publishers say that, without an injunction, Anthropic will keep violating music copyrights through AI training and through Claude outputs that reproduce or imitate protected works.
The lawsuit frames the issue as both a copying problem and a market problem. Publishers allege Claude can reproduce song lyrics verbatim and generate “new” songs that mimic the “heart” of well-known compositions. One example in the complaint describes Claude relying on songbooks or sheet music to respond to prompts asking it to change Eminem lyrics into a song written in “Beyonce’s style.”
The torrenting allegations behind the case
The complaint says Anthropic’s “mass campaign of illegal torrenting” began in July 2021. It alleges that Anthropic co-founder Benjamin Mann “personally used BitTorrent to unlawfully download and upload via torrenting millions of pirated books” from Library Genesis (LibGen). Dario Amodei, Anthropic co-founder and CEO, allegedly approved the torrenting, and both Amodei and Mann are named individually as defendants.
The lawsuit also points to what happened after LibGen was shut down. By the end of 2021, the FBI had shut down LibGen, according to the complaint, but online pirates had already copied its contents into Z-Library. Although Z-Library also got shut down quickly, internal messages revealed in the book authors’ case allegedly showed that Anthropic accessed a copy through the “Pirate Library Mirror” (PiLiMi).
Publishers cite internal Anthropic messages as evidence of the company’s attitude toward those sources. Mann is accused of directing employees to torrent PiLiMi soon after it became available, telling colleagues that the mirror dropped “just in time!” An Anthropic staffer responded, “zlibrary my beloved.”
The music publishers say those comments show that Anthropic celebrated piracy and used torrenting to get training data quickly. They also claim that non-confidential catalogs from LibGen and PiLiMi showed “bibliographic metadata like title, author, and ISBN” indicating that Anthropic torrented at least hundreds of books containing sheet music and song lyrics owned by the suing publishers.
The dispute over what counts as training
Anthropic denies using any books torrented from LibGen and PiLiMi to train commercial Claude AI models, according to the publishers. But the complaint argues that the denial depends on the definition of “training.” Publishers claim that if the court looks far enough back through Anthropic’s AI development process, it may find that commercial Claude models benefited from pirated works indirectly.
The complaint describes a chain in which one AI model may train on “synthetic data” created by another AI model or receive behavioral feedback from another AI model. Publishers allege that Anthropic trained at least one commercially released Claude model using synthetic data created by a non-commercial AI model trained on text derived from LibGen and/or PiLiMi. They also allege that at least one non-commercial model trained on that text provided reinforced feedback to at least one commercial Claude model.
The lawsuit further claims that recently unsealed documents from the book authors’ case showed Anthropic used pirated LibGen books in connection with guardrails. In one filing, an Anthropic witness testified that Anthropic stopped training large language Models (LLMs) on LibGen but continued using the LibGen dataset to check whether long strings of model output too closely matched source text.
Anthropic’s response and the fair use fight
Anthropic pushed back against the publishers’ claims. A spokesperson told Ars that the case was “the third lawsuit from the same lawyers, recycling allegations from cases already before the courts.” The spokesperson added: “Training generative AI models is a transformative fair use—as the court held in Bartz—and we will defend ourselves robustly.”
The publishers’ argument is that the music context may be different from the book authors’ case. The source article notes that Anthropic’s fair use position was helped by book authors’ inability to prove market harms or show that AI tools had substituted them in their markets. Music publishers appear to believe songwriters may have a stronger path on that point.
The complaint says only Anthropic knows how often users rely on Claude to generate substitutes for musicians’ songs. Still, publishers allege that Anthropic publicly tracks the threat to song lyricists from its AI products and knows that AI-generated songs currently topping music charts compete directly with musicians whose songs were allegedly used without payment.
The lawsuit also points to the US Copyright Office, which observed that outputs may dilute royalty pools when generative AI outputs compete in the market for that type of work, including song lyrics, even if they are not substantially similar to a specific copyrighted work.
What the case could clarify
The complaint asks the court to examine several connected questions. Did Anthropic torrent books containing copyrighted lyrics and sheet music? Did those materials influence Claude directly or indirectly? And can AI-generated songs or lyrics cause market harm even when outputs are not exact copies of a protected work?
Publishers also allege that Claude was intentionally trained to regurgitate lyrics, relying on pirate libraries and other unauthorized datasets, including pirated content scraped from lyrics sites without permission from rights holders. They further claim Anthropic destroyed physical books to create unauthorized digital copies “of hundreds of songbooks and sheet music collections.”
The case is therefore about more than one alleged dataset. It is a test of how courts may evaluate AI systems that can produce music-related outputs, especially when the claimed harm is competition with the same creative markets that supplied the training material. For songwriters and publishers, the core allegation is simple: Claude’s development allegedly relied on copyrighted music works without permission, and its outputs may now compete with the people who own those works.