On August 17, 2026, Round Hill Music — an independent music publisher whose catalog of copyrights is valued at roughly $1.1 billion — filed a copyright infringement suit against Anthropic, the AI company behind the Claude chatbot, in the U.S. District Court for the Northern District of California. The case is Round Hill Music LP v. Anthropic PBC, No. 5:26-cv-08505, assigned to Judge Charles R. Breyer. The complaint's core allegation is simple to state and expensive to answer: that Anthropic trained Claude on Round Hill's copyrighted song lyrics without a license, without permission, and by way of pirated copies — and that the bill for doing so could "conceivably exceed $1 billion."
Round Hill did not file alone that day. It filed two complaints simultaneously — one against Anthropic and a companion suit against the AI song generator Suno (which also names the data-scraping firm Bright Data as a co-defendant). The two are built on the same theory but aimed at different targets, and the split tells you something about how music-rights litigation against AI now works.
Two models, two theories
The Suno case covers both compositions and sound recordings, because Suno generates audio — the complaint alleges Suno built its training set by scraping copyrighted music straight off the open web, including YouTube, and used it to power a platform that spits out new songs.
The Anthropic case is narrower by design. Claude is a text model, not an audio generator, so Round Hill's 26-page complaint targets only the lyrics and the underlying compositions. The claim is that Claude ingested copyrighted lyrics without a license and that the material was obtained through the same kind of unauthorized copying and access-control circumvention alleged against Suno. More than 500 of Round Hill's songs are named at the outset — among them "Iris" by the Goo Goo Dolls, "Total Eclipse of the Heart" by Bonnie Tyler, and James Brown's "I Got You (I Feel Good)" — and the filing signals the list could expand to thousands.
To show output-side harm, the complaint points to an experiment: asked to rewrite several of the songs at issue "for a modern audience," Claude allegedly produced text that stayed extremely close to the originals in structure and melodic hooks, changing little beyond surface wording. That is the kind of near-verbatim regurgitation plaintiffs love to cite, because it undercuts the argument that a model has merely "learned from" a work rather than stored and reproduced it.
The number, and the ask
Round Hill is seeking statutory damages of up to $150,000 per work willfully infringed — the math that produces the headline figure of hundreds of millions climbing toward, or past, $1 billion. Stacked on top are Digital Millennium Copyright Act counts: up to $2,500 for each act of circumvention and up to $25,000 for each removal of copyright management information.
But the money may not be the most consequential part of the prayer for relief. Citing 17 U.S.C. § 503, Round Hill asks the court to order the destruction or impoundment of every unauthorized copy of its works — and it defines "copies" to reach training datasets, internal repositories, model weights, and server infrastructure. In plain terms, that is a demand to claw the plaintiffs' songs out of the trained model itself, not just to delete the source files behind it. Each complaint also asks for a full accounting of Anthropic's training data, scraping activity, and datasets involving the works.
That framing is deliberate, and it targets a gap that an earlier settlement left open. When Anthropic settled with a class of book authors for $1.5 billion in September 2025, the deal committed the company to destroying the files it had torrented from shadow libraries — but not the models trained on them. Anthropic certified that no commercially released model had been trained on those datasets. Round Hill's "model weights" language is a bet that the next plaintiff can reach further.
Standing on Bartz v. Anthropic
Round Hill's complaint leans heavily on precedent from the same courthouse. It rests on Bartz v. Anthropic, the authors' case in the Northern District of California, and quotes Judge William Alsup's June 2025 ruling directly: "There is no carveout, however, from the Copyright Act for AI companies." That line is the plaintiff bar's rallying cry — the judicial rejection of the idea that training is categorically different from other copying.
The complaint also revives the piracy narrative that has dogged Anthropic across cases. It sources its account to torrenting from the shadow libraries Library Genesis and Pirate Library Mirror, and quotes what it describes as a co-founder's message to colleagues once the latter could be torrented: "[J]ust in time!" These are allegations, not findings — Anthropic has not answered the complaint, and the fair-use questions at the heart of the dispute remain unresolved.
Why Round Hill says it won't settle
What makes this filing stand out from the pack is the posture. As Suno was reaching settlements with parts of the music industry, Round Hill CEO Josh Gruss signaled the opposite intention: "We intend to take these cases to trial." Round Hill's counsel framed the theory bluntly — that unauthorized training is illegal acquisition, not transformative use, and that there is "nothing fair" about it. For rights holders watching from the sidelines, a plaintiff that publicly refuses to settle is a plaintiff willing to generate the appellate precedent everyone else is waiting on.
The bigger wave
Round Hill is not an outlier; it is the fourth music-industry plaintiff to sue Anthropic, and the suits keep coming. Universal Music Publishing Group, Concord Music, and ABKCO sued the company in Nashville back in October 2023 over roughly 500 songs, a case later transferred to California. The same publishers filed a second, far larger suit in January 2026 covering more than 20,000 works and seeking over $3 billion. BMG brought a third case in March 2026 over 493 compositions. Round Hill's August 17 filing made four.
Then, on August 28, 2026, the publishing arms of Sony Music and Warner Music Group — Sony Music Publishing and Warner Chappell Music — filed a fifth, describing Anthropic's conduct as "one of the largest and most blatant ongoing thefts of intellectual property in history" and naming CEO Dario Amodei and co-founder Benjamin Mann as individual defendants. With that suit, the publishing arms of all three major music companies are now litigating against the maker of Claude at once.
What it means for rights holders
Strip away the celebrity song titles and this is a case about the mechanics of ownership in the AI era — the same questions every catalog owner, songwriter, and independent rights holder now has to think about. Was the work licensed, or scraped? Can a plaintiff force a defendant to purge protected material from a model's weights, or only from a hard drive? And does a training run that never shows the user the underlying work still "use" it in the copyright sense?
The answers will shape how much leverage rights holders actually have when their catalogs turn up inside someone else's machine. Round Hill has bet more than a billion dollars, and its own refusal to settle, on getting those answers in front of a jury. Whatever Judge Breyer and, eventually, the Ninth Circuit decide, the outcome will echo well beyond music — into film, television, and every other corner of the entertainment business where a catalog is the asset.
This post is an editorial explainer based on public court filings and press reporting. All claims described here are allegations unless a court has ruled otherwise; Anthropic has not yet responded to the complaint. Nothing here is legal advice.