Anthropic has another copyright front to fight — and this one comes from the music industry. On Friday, 35 music publishers led by Sony Music Publishing and Warner Chappell Music filed suit in the federal court for the Northern District of California. The accusation is blunt: a “brazen campaign” of intellectual property theft.
At the core is the claim that Anthropic “illegally torrented, scraped and downloaded” copyrighted songs to train the Claude models. The publishers name specific examples, from “Ain’t No Mountain High Enough” to “Eye of the Tiger” to “All I Want for Christmas Is You.” Behind those titles sit the full lyrics and sheet music, not just the names. And the complaint goes a step further: co-founders Dario Amodei and Benjamin Mann are named personally as defendants.
The publishers’ argument reaches past the training itself. They claim Claude can reproduce protected lyrics in its answers — and could therefore stand in for licensed lyric services and undercut the market for human-made music. They’re asking for a jury trial and statutory damages of up to $150,000 per infringed work. Across thousands of compositions that stacks up fast, into the billions, plus claims over stripped copyright-management information.
Anthropic pushes back: it disagrees with the claims and will defend itself robustly.
The conflict isn’t new. Earlier this year Anthropic settled with book authors — for around $1.5 billion, one of the largest copyright settlements of the AI era. So the music publishers have seen that money can flow at the end of the road. And they know their turf: lyrics are a field where rights holders are well organized and quick to litigate.
What interests me about these cases is less the outrage than the underlying question. Where does permitted learning from data end, and where does reproduction begin? A model that pulls patterns from millions of texts is a different thing from a model that spits out a full song lyric on request. That boundary is exactly where these cases get decided — and the outcome shapes how AI companies source their training data from here on. A settlement like the one with the authors is very possible. A ruling that actually draws the line would be far more interesting.
Sources: TechCrunch · Music Business Worldwide