Sony and Warner Sue Anthropic in Multi-Billion-Dollar Music Copyright Case — and Name Dario Amodei Personally

Late on Friday, August 28, some of the world’s largest music publishers walked into the U.S. District Court for the Northern District of California and filed what may become the defining AI copyright battle of 2026. Sony Music Publishing and Warner Chappell Music accuse Anthropic of running “one of the largest and most blatant ongoing thefts of intellectual property in history” — and in an unusual escalation, they named CEO Dario Amodei and co-founder Benjamin Mann as individual defendants, not just the company they run.

The 48-page complaint alleges that Anthropic illegally downloaded and used tens of thousands of copyrighted musical compositions — mostly song lyrics, plus sheet music and derivative works — to train its Claude models. Among the works named in the filing are “Eye of the Tiger,” “All I Want for Christmas Is You,” “Livin’ on a Prayer,” “September,” “Hallelujah,” “Uptown Funk,” and Taylor Swift’s “Paper Rings.” The publishers are seeking a jury trial, statutory damages of up to $150,000 per infringed composition, and up to $25,000 per violation for unlawfully stripping copyright management information from the works. With tens of thousands of works in play, maximum damages could reach several billion dollars.

Anthropic’s response was terse. “We disagree with the publishers’ claims and we intend to defend ourselves robustly in court,” a spokesperson said in a statement.

Why the executives are named personally

The most striking feature of the complaint is its willingness to go after Anthropic’s leadership directly. “Dr. Amodei expressly directed, approved, controlled, and intentionally induced these infringements by Mr. Mann and other Anthropic employees,” the complaint reads. The publishers allege that Mann personally oversaw the torrenting operation and that Amodei knew and approved.

This is not a common strategy in copyright litigation, where companies are usually sued as entities. Naming founders personally raises the stakes dramatically: if the court lets the individual claims proceed past a motion to dismiss, executives across the AI industry could face personal exposure for training-data decisions made years ago. That prospect alone makes this case a landmark, regardless of the final damages figure.

The Bartz precedent: acquisition, not use

To understand why the publishers are confident, look at what happened to Anthropic in the Bartz v. Anthropic case. In August–September 2025, Anthropic agreed to pay $1.5 billion to book authors — the largest copyright settlement in U.S. history. Critically, the judge in that case ruled that using copyrighted books for AI training qualified as fair use, but acquiring those books through illegal BitTorrent downloads from pirate libraries like LibGen was itself infringement. What sank Anthropic was not what the model learned, but how the data got in the door.

The new complaint leans on exactly that weak spot. It alleges Anthropic torrented at least seven million books from LibGen and PiLiMi, treats each act of illegal downloading as a standalone infringement, and argues the analysis holds regardless of whether any particular work ended up inside a commercial Claude model. The music publishers also accuse Anthropic of scraping lyrics from licensed platforms like MusixMatch and LyricFind in violation of their terms of service, and of challenging the legality of widely used datasets such as Books3, The Pile, and Common Crawl.

The synthetic-data loophole argument

Perhaps the most technically interesting claim in the filing concerns synthetic data. Anthropic has publicly denied using LibGen and PiLiMi books to train its commercial Claude models directly. The publishers’ answer: that denial “depends on how Anthropic defines ‘training.’”

According to the complaint, Anthropic trained at least one commercial Claude model on synthetic data generated by a non-commercial model that had itself learned from the pirated LibGen and PiLiMi texts, and also used such a model to provide reinforcement-learning feedback to a commercial model. If proven, this would establish a kind of “laundering” theory of copyright liability — that distilled or synthetic data derived from pirated sources carries the original taint. Courts have never squarely addressed this question, and the answer could reshape how every AI lab builds training pipelines. The full scope of these claims is expected to surface during discovery.

A closing international window

The case does not sit in isolation. In November 2025, the Munich Regional Court ruled in a case involving GEMA that copyright-protected song lyrics count as reproductions even when stored within a model’s parameters, and that a chatbot outputting those lyrics amounts to unlawful public disclosure. The German court held the model operator — not the user who typed the prompt — responsible, even for prompts deliberately engineered to elicit copyrighted lyrics. With Europe constraining output-side liability and the U.S. case now attacking the input side, the space for “we just trained on what was available” defenses is narrowing from both directions.

What this means for the industry

Three implications stand out. First, the Bartz settlement established that piracy-based acquisition is the fatal flaw in AI copyright defenses, and every publisher with a catalog now has a playbook. Second, the personal-liability claims against Amodei and Mann, if allowed to proceed, will change how AI executives document and approve data-sourcing decisions — expect general counsel to become far more involved in training pipelines. Third, the synthetic-data theory threatens the industry’s favorite workaround: if courts accept that models trained on pirated data pass their legal defects to downstream distillations, the provenance of every synthetic dataset becomes a litigation risk.

For Anthropic, the timing is awkward. The company has positioned itself as the responsible, safety-first AI lab — a brand built partly on contrast with more aggressive competitors. A complaint describing “a brazen campaign of illegally torrenting, scraping, and downloading copyrighted works,” aimed at its CEO personally, cuts directly against that narrative. Some of the same lawyers representing Sony and Warner also represent Concord Music Group and Universal Music Group in a similar case filed in January, which suggests a coordinated, multi-front campaign by the music industry rather than a one-off dispute.

The publishers’ framing is aggressive, but the underlying legal theory has already won once. Anthropic will presumably fight the individual-defendant claims and the synthetic-data theory hard, and discovery in this case — covering how the LibGen and PiLiMi datasets actually flowed into Claude’s training runs — will be among the most closely watched in the tech industry. What emerges will determine whether the $1.5 billion Bartz settlement was the ceiling for AI copyright exposure, or merely the first installment.