Sony, Warner Music Sue Anthropic Over Songs Used in AI Training

Anthropic agreed to pay $1.5 billion in a recent copyright case, yet Sony Music Publishing and Warner Chappell Music claim the settlement does not fully resolve their grievances. The publishers recently filed a new lawsuit in a California federal court, specifically targeting Anthropic's CEO Dario Amodei and co-founder Benjamin Mann.

The complaint alleges that Anthropic engaged in a "blatant violation of copyright law" by unlawfully downloading, scraping, and torrenting thousands of protected musical works from unauthorized sources. These sources, according to the publishers, included Library Genesis, the Pirate Library Mirror, and licensed lyrics platforms like Musixmatch and LyricFind.

Anthropic allegedly gathered songs from these sites, which subsequently helped train their AI model, Claude. The music publishers assert that Claude can reproduce protected lyrics in its responses, citing famous works such as "Hallelujah," "Uptown Funk," "All I Want for Christmas Is You," Taylor Swift's "Cruel Summer," and "Paper Rings."

The earlier settlement covered claims about specific books and Anthropic's past acquisition of them, but it did not provide a general license for every copyrighted work, nor did it settle unrelated music publishers' claims. Sony and Warner argue the payment was merely a business cost for the valuable company and seek up to $150,000 for every work wilfully infringed, as well as $25,000 for each instance copyright-management information was allegedly removed from protected material.

With thousands of songs involved, potential damages could reach several billion dollars. The publishers must still prove ownership, copying, infringement, and any claimed wilfulness. Sony and Warner believe their claims remain unpaid and that Anthropic continued using an unlawful sourcing model, making the $1.5 billion settlement insufficient.

The court will now decide whether these accusations survive evidence and legal scrutiny, a decision that may influence future agreements between creators and developers. Anthropic rejects the allegations and vows to defend itself robustly, asserting that training generative AI models is a transformative fair use. The dispute concerns not only whether machines may learn from creative works but also where the training material originated and whether outputs reproduce protected expressions too closely.