Late last Friday evening (August 28), Sony Music Publishing and Warner Chappell Music, leading a coalition of music publishers, filed suit in the U.S. District Court for the Northern District of California. Named as defendants are Anthropic — the company behind Claude — and two of its co-founders, Dario Amodei and Benjamin Mann. The complaint is unusually harsh in its language, with plaintiffs calling the conduct "one of the largest, most egregious, sustained thefts of intellectual property in history." Sony Music Publishing is a unit of Sony Group, while Warner Chappell is the music publishing arm of Warner Music Group; combined, the plaintiffs hold a formidable catalog — which is why the case has been seen as heavy-hitting from the start.
The Core Allegation: A "Brazen" Piracy Campaign
The 48-page complaint alleges that Anthropic carried out "a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works" — using BitTorrent seeding and web scraping to acquire, without authorization, "tens of thousands" of copyrighted musical works (including lyrics and sheet music) to train its Claude models. The complaint cites specific works as examples, including "Ain't No Mountain High Enough," "All I Want for Christmas Is You," "Eye of the Tiger," "Here Comes Santa Claus," and "Paper Rings" — hardly obscure tracks, but songs familiar to the general public, underscoring the scale of the alleged infringement. The complaint further alleges these works were "reproduced multiple times," both as training inputs and in content generated by the models.
According to the plaintiffs, the acquisition itself was notably aggressive: between 2021 and 2022, Anthropic allegedly used BitTorrent to download more than 7 million titles from pirate "shadow libraries," a trove interlaced with large amounts of lyrics and sheet music; it also directly scraped data from lyrics sites such as MusixMatch and LyricFind.
Claims and Demands: Up to $150,000 Per Work, Potential Billions
In terms of claims, the plaintiffs assert four causes of action: (1) direct infringement for torrenting, against Anthropic, Amodei, and Mann; (2) inducement of infringement against Amodei and Mann personally; and (3)–(4) additional infringement claims directed at Anthropic itself. The plaintiffs seek a jury trial and set out two damages yardsticks: up to $150,000 per infringed work (the statutory maximum for willful infringement), plus an additional $25,000 for each removal of copyright management information (CMI).
On that basis, with "tens of thousands" of works at issue, Anthropic's potential exposure could reach into the billions of dollars. Beyond monetary damages, the plaintiffs ask the court to order the destruction of all infringing copies and to require Anthropic to provide a full accounting of Claude's training data. The case is being handled by Oppenheim + Zebrak and Pryor Cashman — the former being the lead firm in both the Concord/UMG cases and the Bartz case.
As of press time, Anthropic has responded publicly. A company spokesperson said in a statement: "We disagree with the publishers' claims and we intend to defend ourselves robustly in court."
A String of Prior Cases: Anthropic's Copyright Litigation History
Stepping back, this is far from the first time Anthropic has been sued over copyright.
In January of this year, Universal Music Group and Concord Music Group already sued Anthropic, alleging the illegal download of more than 20,000 copyrighted songs for AI training and seeking up to $3 billion. The same Oppenheim + Zebrak team is involved, and the two cases are cut from the same cloth in approach and evidence chains. Earlier still, in October 2023, UMG, Concord, and ABKCO brought their first suit against Anthropic, seeking roughly $75 million over Claude reproducing lyrics. Separately, music rights manager BMG has also sued Anthropic (foreign reports say the case centers on about 493 works); and just last week, independent music publisher Round Hill Music sued both Suno and Anthropic, seeking upward of $1 billion in each case and explicitly declining to settle.
The case that truly set the tone for everything that followed is Bartz v. Anthropic, a class action brought by authors. In June 2025, the court issued a pivotal ruling: training an AI model on copyrighted works may itself qualify as fair use, but acquiring those works through pirated channels — and retaining them long-term — does not. In September 2025, Anthropic settled the piracy portion with the plaintiffs, agreeing to pay at least $1.5 billion, formally approved by the San Francisco federal court in July 2026 — one of the largest copyright settlements in U.S. history. The subtext of that precedent is clear: the Sony/Warner, BMG, and Round Hill cases are all really about the same point — not that using copyrighted content for training is illegal, but that how the content was obtained is where the line was crossed.
What's Different About This Case: Shifting the Focus from "Model Output" to "Data Sourcing"
Compared with earlier cases, this complaint is broader and focuses more squarely on the act of piracy itself. It specifically alleges that Anthropic used illegal torrenting to obtain millions of text copies — including not just books but content containing lyrics and sheet music. That means the evidentiary chain may reach well beyond the surface question of "the model reproduced copyrighted content" and back into Anthropic's internal data collection, cleaning, and retention processes — precisely why the plaintiffs demand a full description of the training data.
This is also why the case is seen as potentially far-reaching: it shifts the focus from "what the AI model generated" to "where an AI company's training data actually came from." For Anthropic, the legal risk ahead is likely to concentrate on data procurement and internal data provenance — not just on what Claude says in conversation. For the industry as a whole, the argument "we didn't use these data to train our current models" may no longer be a defensible compliance answer. If courts start digging into intermediate models, synthetic data, and feedback pipelines, data traceability will become an issue no frontier lab can sidestep — and this case could become a landmark reference.
The case is still in its early stages, and Anthropic has said it will contest it vigorously. Whether it ends in a mega-settlement like Bartz, or drags into a longer war of attrition, is worth watching closely.