Anthropic Sued Over Copyrighted Music Used to Train Claude

Sony Music Publishing, Warner Chappell Music and affiliated publishers sued Anthropic over allegations that copyrighted music was illegally obtained and used to develop Claude

Anthropic already paid $1.5 billion to settle claims that it pirated books to train Claude. Now the music industry is coming after the company with a similar allegation, but with a sharper legal argument: the problem wasn't simply what Claude learned. It was how Anthropic allegedly got the material in the first place.

Sony Music Publishing and Warner Chappell Music, joined by numerous affiliated publishers, sued Anthropic late Friday in the U.S. District Court for the Northern District of California.

"Sony Music Publishing and Warner Chappell Music have filed a lawsuit against Anthropic, alleging that the company engaged in a 'brazen campaign of illegally torrenting, scraping, and downloading copyrighted works," Music Business Worldwide reported after obtaining the 48-page complaint.

Anthropic has not publicly responded to this specific complaint. TechCrunch reported the company could not be reached for comment before publication. But the lawsuit arrives with unusually convenient timing for the plaintiffs: a federal court has already ruled in Anthropic's books case that acquiring copyrighted works through piracy is a separate problem from whether those works can legally be used to train an AI model.

"While training an AI model on copyrighted works can be fair use, that does not mean that Anthropic's acquisition of those works was lawful," U.S. District Judge William Alsup wrote in Bartz v. Anthropic, finding that Anthropic's downloading and storage of pirated books was not protected by fair use.

The Fight Is About Acquisition

In Bartz v. Anthropic, a judge found that using copyrighted books to train AI could, under certain circumstances, qualify as fair use. But that did not make the way Anthropic obtained those books legal. The court found that the company had acquired millions of works from pirate sources including LibGen and the Pirate Library Mirror, describing the conduct as "straightforward piracy but at massive scale."

Anthropic ultimately agreed to a $1.5 billion settlement, which received final approval in July 2026. "The $1.5 billion settlement provides meaningful relief to the Class," U.S. District Judge Araceli Martínez-Olguín wrote when granting final approval of the settlement in July 2026.

The music publishers are building directly on that precedent. Their argument is effectively that even if Anthropic eventually wins some fair-use protection for training Claude on copyrighted lyrics, that protection cannot excuse allegedly torrenting and downloading the works in the first place.

That gives the case a much narrower question than the sweeping debate over whether AI companies can train on copyrighted material. The issue is whether the acquisition pipeline itself violated copyright law.

DimensionDetails & ParametersPlaintiffsSony Music Publishing, Warner Chappell Music, and affiliated music publishersDefendantsAnthropic PBC, Dario Amodei (CEO & Co-founder), Benjamin Mann (Co-founder)JurisdictionU.S. District Court for the Northern District of CaliforniaStatutory Relief SoughtUp to $150,000 per work for willful infringement; up to $25,000 per violation for removal of Copyright Management Information (CMI)Legal Precedent LeverageBartz v. Anthropic (approved $1.5 billion settlement over pirated book corpora)

What The Publishers Want

The complaint alleges direct copyright infringement through the alleged torrenting activity against Anthropic, Amodei and Mann. It also accuses Amodei and Mann personally of contributory infringement for allegedly directing and approving the activity.

"Anthropic's infringement threatens to substitute Claude-generated lyrics for the original works, thereby depriving Plaintiffs and other music publishers and songwriters of licensing opportunities and revenue," the publishers allege in their complaint filed in the Northern District of California.

The complaint names "tens of thousands" of compositions, including Ain't No Mountain High Enough, All I Want for Christmas Is You, Eye of the Tiger, Livin' On a Prayer, September, Hallelujah and Taylor Swift's Paper Rings.

The plaintiffs are seeking statutory damages of up to $150,000 per willfully infringed work, along with as much as $25,000 for each alleged removal or alteration of copyright-management information. They are also asking for an accounting of Claude's training data and destruction of infringing copies.

Claude's Guardrails Face A Test

Anthropic added safeguards following earlier litigation designed to stop users from getting the model to reproduce protected material. The new complaint argues those protections can be circumvented simply by changing or repeating prompts.

That matters because the publishers are not arguing only that copyrighted material went into the training system. They are pointing to what Claude can produce and arguing that the resulting outputs can compete with the works Anthropic allegedly used to build the model.

Whether that argument survives in court is another matter. The complaint establishes what the publishers allege, not what a judge has found.

The Music Cases Keep Multiplying

Universal Music Publishing Group, Concord Music Group and ABKCO sued the company in 2023 over roughly 500 songs. The case was later transferred to California, and a January 2026 lawsuit from the same publishers expanded the fight to more than 20,000 works and sought more than $3 billion.

BMG filed another case in March 2026 covering hundreds of compositions. Round Hill Music followed with another lawsuit on August 17.

The latest filing puts the publishing arms of all three major music companies into litigation against Anthropic. What looks like repeated litigation over the same basic technology is partly a consequence of how music rights are divided: lyrics, musical compositions and sound recordings can carry separate copyrights controlled by different owners.

One song can therefore produce multiple potential plaintiffs, each with a different legal interest in what an AI company allegedly copied.

The Rule That Could Spread

If courts continue separating how copyrighted material was acquired from whether that material could legally be used for AI training, then an AI company's data pipeline becomes a legal issue in its own right.

That would matter to every AI lab that built training datasets from material gathered across the open internet, unofficial repositories or other sources whose copyright status was uncertain. A company might eventually establish that a particular use of copyrighted material was transformative or otherwise protected.

The music publishers are betting that distinction gives them a stronger case than simply arguing that Claude learned from copyrighted lyrics.

And that may be the most consequential part of this lawsuit. The fight isn't only over what AI is allowed to learn. It is increasingly about whether AI companies were legally entitled to obtain the material they taught it in the first place.