On August 29, Sony Music Publishing and Warner Chappell Music filed a federal lawsuit against Anthropic in the Northern District of California. That alone would be significant. What makes the filing unusual is who else is named on the complaint: not just the company, but CEO Dario Amodei and cofounder Benjamin Mann, sued as individuals.
Naming founders personally is a signal, not a formality. Plaintiffs use it when they want defendants to feel a case in a way a corporate entity cannot, and when they believe the underlying conduct was decided at the top rather than buried in an engineering team’s data pipeline. Sony and Warner are telling the court, and Anthropic’s leadership, that they see this as a choice someone made, not an accident.
The lawsuit in brief
- Filed August 29 in the U.S. District Court for the Northern District of California
- Plaintiffs: Sony Music Publishing and Warner Chappell Music
- Individually named: CEO Dario Amodei and cofounder Benjamin Mann
- Alleges torrenting, scraping and downloading of “tens of thousands” of copyrighted compositions
- Cites Common Crawl, Library Genesis and Pirate Library Mirror as sources
- Seeks up to $150,000 per work willfully infringed, plus up to $25,000 per removed copyright notice
What the complaint actually alleges
The filing describes what the publishers call a “brazen campaign” of illegally torrenting, scraping and downloading copyrighted works at scale. That framing matters, because it is narrower than “AI training is theft,” the slogan version of this fight that has circulated since 2023. The complaint is not arguing that training a model on copyrighted lyrics is inherently illegal. It is arguing that Anthropic got the lyrics through channels that were illegal regardless of what happened next.
According to the complaint, Anthropic pulled lyrics and sheet music from Common Crawl’s web archive, and separately from Library Genesis and Pirate Library Mirror, two shadow libraries built from pirated books and texts. Those same two shadow libraries were at the center of Anthropic’s previous, much larger legal exposure over training data, and that history is not a coincidence in how this case is built.
The publishers also allege something beyond training: that Claude will reproduce lyrics verbatim when a user asks for them. If accurate, that turns the dispute from a purely upstream question about what went into the model into a downstream question about what comes back out, which is generally an easier claim for a copyright holder to prove.
Why the Bartz settlement makes this case different
To understand why Sony and Warner built the complaint this way, it helps to know what already happened to Anthropic once. In September 2025, the company settled a class action brought by authors over pirated training data for $1.5 billion, one of the largest copyright settlements on record. The judge in that case, Bartz v. Anthropic, made a specific and widely quoted distinction: training a model on copyrighted text can qualify as transformative fair use, but obtaining that text through piracy is a separate legal problem that fair use does not excuse.
That distinction is now doing the heavy lifting in the music case. Sony and Warner are not primarily fighting about whether Claude learning from song lyrics is fair use in the abstract. They are arguing that Library Genesis and Pirate Library Mirror, the same pirate sources named in Bartz, are illegitimate no matter what legal theory covers the training step that follows. It is the same playbook that produced a $1.5 billion outcome once already, now aimed at a different category of copyrighted work with a different, and in some ways sharper, damages formula.
| Detail | Bartz v. Anthropic (2025) | Sony/Warner v. Anthropic (2026) |
|---|---|---|
| Plaintiffs | A class of book authors | Sony Music Publishing, Warner Chappell Music |
| Named individuals | Anthropic only | Dario Amodei and Benjamin Mann, plus Anthropic |
| Alleged sourcing | Pirated books, including Library Genesis | Common Crawl, Library Genesis, Pirate Library Mirror |
| Outcome sought | Settled for $1.5 billion | Pending, up to $150,000 per work |
“Training generative AI models is a transformative fair use, as the court held in Bartz, and we will defend ourselves robustly.”
Anthropic, statement responding to the lawsuit
Notice what that statement does and does not address. It reasserts the fair use argument that worked, in part, for Anthropic in Bartz. It says nothing about how the lyrics were obtained in the first place, which is precisely the part of the earlier ruling that went against the company and precisely the part Sony and Warner built their new complaint around.
Sony Music Publishing and Warner Chappell control catalogs spanning decades of recorded music, the same catalogs the complaint says ended up in Claude’s training data through pirate archives. Photo via Pexels.
The math behind $150,000 a song
The number in the headline comes from the U.S. Copyright Act’s provision for statutory damages, which allows a court to award up to $150,000 per work when the infringement is found to be willful, without the plaintiff having to prove actual financial harm for each individual song. The complaint also seeks up to $25,000 for every instance in which copyright management information, the metadata identifying a song’s writer and publisher, was allegedly stripped out during scraping.
Why the exposure gets so large so fast
Figures reflect the statutory maximum named in the complaint. A court would set the actual per-song award, and it is rarely the maximum for every work in a large catalog case.
That last caveat matters more than the headline number does. Courts almost never apply the statutory ceiling uniformly across tens of thousands of works in a single case, and Anthropic will argue for a lower per-song figure, a smaller count of provably infringed works, or both. What the $150,000 figure actually does is set the outer boundary of Anthropic’s exposure and its incentive to settle rather than litigate a catalog this large to a verdict.
What it means beyond Anthropic
This case lands on top of a legal landscape that copyright holders have already tested extensively against AI companies this year. GeekBlog covered the record-setting copyright class action certified against the AI industry and, separately, Anthropic’s own lawsuit against the federal government over a contracting blacklist, so the company is currently fighting legal battles on more than one front at once, with regulators on one side and rights holders on the other.
For the music industry, the strategic logic is straightforward: Bartz proved that a coordinated, well-funded lawsuit built around piracy sourcing rather than pure fair use theory can produce a settlement in the billions. Sony and Warner are running the same play, and every other publisher and label watching this case is learning whether it works twice. For Anthropic, the timing is awkward. The company has spent 2026 positioning Claude’s technical capabilities as proof it belongs at the frontier of the industry, and a second nine-figure or ten-figure copyright settlement would complicate that story regardless of how the underlying technology performs.
