Sony Music Publishing, Warner Chappell Music and dozens of affiliated publishers have sued Anthropic over material allegedly used to build Claude, accusing the AI company of illegally obtaining and copying copyrighted lyrics, sheet music and musical compositions for training.
The lawsuit was filed August 28th in the U.S. District Court for the Northern District of California and names Anthropic CEO Dario Amodei and co-founder Benjamin Mann as individual defendants alongside the company. The publishers allege Anthropic obtained copyrighted works through torrenting, web scraping, third-party datasets and other methods without permission or payment.
The complaint covers music associated with artists and songwriters ranging from The Beatles and Michael Jackson to Taylor Swift, Mariah Carey and Bon Jovi. Music Business Worldwide places the filing inside a widening legal battle between major music publishers and Anthropic over the material used to develop Claude.
Anthropic disputes the allegations. “We disagree with the publishers’ claims,” a company spokesperson told TechCrunch, adding that the company intends to defend itself.
The case reaches beyond the increasingly familiar question of whether copyrighted work can be used to train generative AI. Sony and Warner are also challenging how Anthropic allegedly obtained the material in the first place, putting piracy, training copies, Claude’s outputs and music-publishing rights inside the same dispute.
The Lawsuit Targets Songs, Not Master Recordings
The rights at issue are important because “music copyright” can describe different assets.
A commercially released song commonly involves at least two major copyright layers. One covers the sound recording, often controlled by a label or artist. Another covers the underlying composition, including lyrics and musical notation, typically administered by songwriters and music publishers.
The new Anthropic lawsuit comes from the publishing side.
Sony Music Publishing and Warner Chappell are not alleging simply that Claude copied finished commercial recordings by famous performers. Their claims concern copyrighted musical compositions contained in lyrics, sheet music, books, websites and other material Anthropic allegedly copied while developing its AI systems.
Reuters identifies works associated with The Beatles, Taylor Swift and Michael Jackson among the material cited in the case.
That distinction separates the lawsuit from cases aimed at generative music services over alleged copying of commercial sound recordings. A composition can be protected independently of any particular recording of it.
For publishers and songwriters, lyrics and written music are therefore training material in their own right rather than incidental information surrounding a master recording.
The Publishers Say Anthropic Pirated Training Material
The sharpest allegation concerns acquisition.
Sony and Warner’s publishers accuse Anthropic of using torrenting, shadow libraries, web scraping and other methods to assemble copyrighted material without licenses. The allegations connect the music case to earlier disputes over large collections of pirated books used in AI development.
Among the sources identified in reporting on Anthropic’s training libraries are shadow libraries including Library Genesis and Pirate Library Mirror. Collections obtained through those services allegedly included books containing copyrighted lyrics and musical notation controlled by publishers.
Other material was allegedly acquired through scraped websites, third-party datasets and digitized physical works.
The publishers’ theory separates two acts that can easily become blurred in AI copyright debates: obtaining a copyrighted copy and subsequently using that copy for model development.
That distinction has already mattered in litigation involving authors. TechCrunch’s legal overview notes that earlier proceedings distinguished the question of AI training from the separate issue of obtaining books through piracy.
The music publishers are now attempting to put that acquisition question directly at the center of a dispute over songs.
Amodei and Mann are personally named as part of that theory. The publishers allege Mann participated in obtaining pirated material and that Amodei was involved in directing the conduct. Those are allegations in the complaint, not findings of personal liability.
Naming company executives makes the filing more specific than a claim that copyrighted works simply appeared somewhere inside a vast dataset. The publishers are alleging identifiable decisions about how data was collected and retained.
Claude’s Outputs Are Part of the Case Too
The claims do not stop at Anthropic’s training library.
Sony and Warner also allege Claude can reproduce protected lyrics verbatim and generate material substantially similar to copyrighted compositions. That gives the lawsuit another layer beyond what happened before a model was trained.
Reuters’ review of the filing says the publishers accuse Claude of reproducing lyrics and generating material that can compete with the original works.
The complaint also raises claims involving copyright management information under the Digital Millennium Copyright Act. Publishers allege identifying or ownership information connected to protected works was removed or altered during Anthropic’s copying process.
The case therefore presents several legal theories rather than one abstract claim called “AI copyright”: allegedly unlawful acquisition, copying for model development, potentially infringing outputs and treatment of copyright-management information.
Anthropic disputes the publishers’ case, and none of those allegations has been established as fact by a court.
All Three Major Publishing Groups Are Fighting Anthropic
Sony Music Publishing and Warner Chappell are entering a legal fight that already includes Universal Music Publishing Group and other publishers.
Universal, Concord Music Group and ABKCO first sued Anthropic in 2023 over Claude’s alleged reproduction of copyrighted lyrics. That dispute initially involved roughly 500 works and continued through amended complaints and procedural fights.
A later case expanded the conflict dramatically, adding allegations over Anthropic’s acquisition of copyrighted material through torrenting and seeking substantial damages over a much larger group of works.
With Sony and Warner publishing companies now pursuing their own claims, publishing operations associated with all three major music groups are involved in litigation against Anthropic.
MBW’s coverage places the August 28th filing inside that broader campaign rather than treating it as the music industry’s first confrontation with Claude.
The lawsuits are not interchangeable. They involve different plaintiffs, catalogs, works and allegations, and each case has its own procedural history.
Together, however, they show music publishers pressing courts on two related questions: whether copyrighted compositions can be copied for generative AI training without licenses and whether the source of those training copies changes the legal analysis.
Fair Use May Not Answer the Piracy Question
Fair use remains central to AI copyright litigation, but Anthropic’s legal battles demonstrate why it may not answer every part of the dispute.
AI developers have argued that model training can be transformative because systems analyze large collections of material to identify patterns rather than distributing ordinary substitute copies of each work to consumers. Copyright owners argue that copying protected material at scale without licenses deprives creators and rightsholders of control and can contribute to systems capable of reproducing or competing with their work.
The Anthropic cases introduce another question before that debate is resolved: where did the training copy come from?
In earlier litigation involving authors, a federal judge distinguished between using copyrighted books for AI training and maintaining a library assembled from pirated books. That distinction gives Sony and Warner a legal strategy that does not depend entirely on convincing a court that every act of AI training is inherently infringing.
The publishers can separately challenge alleged torrenting and other acquisition methods.
Whether they can prove those allegations for the musical compositions in this case remains for the court to decide.
The Potential Damages Reach Into the Billions
The scale of the requested relief reflects the number of copyrighted works the publishers say are involved.
U.S. copyright law can permit statutory damages of as much as $150,000 for each work found to have been willfully infringed. The complaint also seeks additional relief connected to alleged violations involving copyright management information.
Those maximum statutory figures should not be mistaken for money Anthropic has already been ordered to pay. No damages have been awarded, and the publishers still have to establish their claims and entitlement to relief.
The potential exposure nevertheless becomes enormous when damages can be calculated work by work across a large music catalog, which is why the dispute has been described as potentially multibillion-dollar litigation.
Publishers are also seeking injunctive relief that could affect Anthropic’s future use of protected material, along with remedies concerning allegedly infringing copies and information about training data.
For the AI business, those requests matter beyond a single damages calculation. A ruling that restricts how copyrighted compositions can be acquired or retained for training could influence how future datasets are assembled before a model is ever released.
Music Publishing Could Push AI Toward Licensing
The Anthropic music lawsuit arrives while rightsholders across books, journalism, visual art and other creative fields are testing how existing copyright law applies to generative AI.
Music publishing brings a particularly mature licensing system into that argument. Publishers already administer permissions and payments across streaming, reproduction, synchronization, public performance and numerous other uses of compositions.
AI training creates a different kind of demand for enormous catalogs of material.
If courts conclude that some forms of training require authorization, or that companies cannot build training libraries from unlawfully obtained copies, AI developers could face growing pressure to negotiate scalable licenses with publishers before using large song catalogs.
That outcome is not predetermined by the Sony and Warner case. Anthropic disputes the allegations, and courts are still developing rules around different kinds of AI copying and training.
What the filing does establish is how aggressively major music publishers are trying to shape those rules before industry practices harden around them.
Sony Music Publishing and Warner Chappell are effectively arguing that building Claude does not create a copyright-free zone around musical compositions, particularly when the copies themselves were allegedly obtained from unauthorized sources.
Anthropic will have the opportunity to contest both the factual allegations and the publishers’ legal theories.
The immediate question is therefore narrower than whether artificial intelligence and music can coexist. Courts will have to examine what Anthropic actually copied, where those copies came from, how they were used, what Claude can reproduce and which copyright rules apply at each stage.
For music publishers, that distinction could be decisive. The fight is no longer only about whether an AI model has encountered a copyrighted song.
It is increasingly about how that song entered the machine.