Anthropic Sony piracy lawsuit cites staff chat about Z-Library
Music publishers allege Anthropic obtained pirated books containing lyrics and sheet music; Anthropic denies the books trained commercial Claude models.
By Dominic Okoye · Staff Writer
· 3 min read
Music publishers including Sony, EMI and Warner Chappell have filed an Anthropic Sony piracy lawsuit that alleges the AI company acquired pirated books containing lyrics and sheet music tied to their copyrighted compositions. The complaint uses internal staff messages about the Pirate Library Mirror, including one employee’s “zlibrary my beloved” reply, to argue that Anthropic embraced torrenting as a way to obtain data.
The case puts a pointed internal exchange alongside a broader, unresolved question for AI copyright litigation: whether allegedly infringing material reached commercial models, directly or through intermediate systems. The complaint’s allegations have not been adjudicated.
What does the Z-Library chat allege about Anthropic?
According to Ars Technica’s account of the complaint, Anthropic co-founder Benjamin Mann told colleagues that the Pirate Library Mirror, known as PiLiMi, had arrived “just in time!” The mirror was a copy of material associated with Z-Library, which itself had copied Library Genesis, or LibGen. Another staff member responded, “zlibrary my beloved.”
The publishers contend that the exchange helps show an internal willingness to use pirate-library sources and torrenting to obtain training material quickly. They allege that searches of LibGen and PiLiMi catalog data identified at least hundreds of books containing sheet music or lyrics for compositions they control. The complaint also names Mann and Anthropic CEO Dario Amodei as defendants, Ars Technica reported.
A chat message is not, by itself, proof of infringement or a complete record of a model’s training corpus. The litigation instead turns on the publishers’ allegations about acquisition, use and market harm, as well as Anthropic’s defenses.
Were the allegedly pirated books used to train Claude?
Anthropic has denied using books torrented from LibGen and PiLiMi to train its commercial Claude models, according to Ars Technica. The publishers argue that the distinction may not settle the issue. Their theory is that at least one non-commercial model trained on text derived from those sources may have created synthetic data for a commercial Claude model, or supplied reinforcement feedback used in commercial-model development.
That is an allegation about an indirect development pipeline, rather than an established account of Claude’s training. Ars Technica also reported that, in separate authors’ litigation, an Anthropic witness said the company had stopped training large language models on LibGen but continued to use the dataset to check whether long output passages too closely matched source text.
Anthropic said the action was the third lawsuit brought by the same lawyers and characterized it as recycling allegations already before courts. A company spokesperson said training generative AI models is transformative fair use and that Anthropic would defend the case.
The publishers also pointed to Anthropic’s prior $1.5 billion settlement with book authors, which Ars Technica reported concerned allegations over pirated books. The new suit extends the fight into music publishing, where plaintiffs will need to establish their own claims concerning the works at issue and any use connected to Anthropic’s products.
This story draws on original reporting from Ars Technica.
