Sony and Warner Chappell Sue Anthropic Over Song Lyrics

Sony Music Publishing and Warner Chappell Music filed a copyright lawsuit against Anthropic on August 28, and the complaint is notable less for the allegation than for who it names. Alongside Anthropic as a company, the 48-page filing lists co-founders Dario Amodei and Benjamin Mann as individual defendants. That’s a different kind of legal exposure than a corporate defendant faces, and it signals the publishers think they can prove the piracy was a decision made by specific people, not an ambient policy failure.

The core claim is familiar from a year of AI copyright litigation: Anthropic trained Claude on tens of thousands of copyrighted compositions, obtained through what the complaint calls torrenting, scraping, and downloading pirated collections, without a license. The songs named in the suit range from Mariah Carey’s “All I Want for Christmas Is You” to Taylor Swift’s “Paper Rings” to Marvin Gaye’s “Ain’t No Mountain High Enough.” The publishers are seeking a jury trial and statutory damages of up to $150,000 per infringed work, which at “tens of thousands” of compositions is a headline number that could run past a billion dollars before a jury ever weighs in.

This isn’t a new legal theory — it’s the same one that already cost Anthropic $1.5 billion

In September 2025, Anthropic settled a class-action suit brought by authors over the same underlying practice: training on books obtained through piracy. That settlement followed a judicial finding that cuts closer to the center of this new case than it might first appear. The court in the authors’ case held that training an AI model on copyrighted material isn’t, by itself, illegal — but acquiring that material through piracy is. Anthropic paid $1.5 billion on the strength of that distinction.

Sony and Warner Chappell are suing on exactly the same theory, aimed at music instead of books. If the acquisition method is what matters legally, rather than the training itself, this suit inherits most of its argument from a fight Anthropic already lost. Anthropic’s public response — “we disagree with the publishers’ claims and we intend to defend ourselves robustly in court” — doesn’t preview a different legal strategy, and there isn’t an obvious one available. The more interesting fight may end up being about numbers: how many songs, obtained which way, at what per-work damage figure.

Why naming Amodei and Mann personally is the real story

Corporate defendants pay fines that get budgeted for. Individual defendants face something else: a claim that they knew, and did it anyway. Naming founders personally is a standard-but-aggressive litigation tactic, usually reserved for cases where a plaintiff believes the internal record — emails, decisions, sign-offs — ties specific people to specific choices. It also raises the settlement stakes in a different way than a corporate-only suit would, since individual defendants have their own incentives to see a case resolved rather than litigated to discovery.

That pattern of internal records deciding a case is one this blog has now seen twice in two weeks. The Pentagon blacklist ruling against Anthropic turned entirely on a paper trail — internal emails that contradicted the government’s official story enough for a federal judge to call the stated rationale “illegal and baseless.” Litigation increasingly runs on what a company’s own communications say happened, not on what a company says happened after the fact. Whether Sony and Warner Chappell have that kind of record for Amodei and Mann specifically is the detail that will decide whether the personal-liability claim survives a motion to dismiss or turns out to be a pressure tactic aimed at a faster settlement.

The bigger pattern: music is now fully in the AI copyright fight

Music publishers moving against a frontier lab isn’t new in the abstract — Universal Music Group and others sued Suno and Udio last year over AI-generated songs that competed directly with catalog music. This case is different in kind: Claude doesn’t generate music, and Anthropic isn’t accused of building a competing product. The claim here is purely about training data, which puts it in the same bucket as the book publishers’ and authors’ suits rather than the music-generation suits. That distinction matters for how the case gets argued, but it also means every major category of copyrighted media — books, images, code, and now song lyrics — has an active, well-funded plaintiff suing a frontier AI lab over how its training data was sourced.

For Anthropic, that adds up to a straightforward pattern regardless of how this specific suit resolves: the $1.5 billion book settlement didn’t close the company’s copyright exposure, it set a price and a legal template that other rightsholders are now using. Whether music ends up costing more, less, or about the same per work than books did will be one of the more closely watched numbers in AI litigation over the next year.

Sources: TechCrunch, Axios, Music Business Worldwide, Engadget, Business Standard.