Music publishers sue Anthropic for $3 billion over pirated songs

The complaint names Dario Amodei and Benjamin Mann as individual defendants alongside the company. According to the filing, evidence uncovered in the Bartz v. Anthropic case showed that Mann 'personally used BitTorrent to download via torrenting from LibGen approximately five million copies of pirated books' for Anthropic's use in June 2021. The complaint further alleges that Amodei 'personally discussed and authorized this illegal torrenting' despite internal warnings that LibGen was described internally as 'sketchy.'

The lawsuit grew out of an earlier case — Concord Music Group v. Anthropic (Concord I) — filed in October 2023 over 499 musical compositions, with potential damages of roughly $75 million. Discovery in a separate case, Bartz v. Anthropic, revealed in July 2025 that the torrenting had swept up thousands of additional works. The publishers tried to add the piracy claims to Concord I but the court denied that motion in October 2025, ruling they had failed to investigate the piracy claims earlier — forcing the separate, dramatically expanded January 2026 filing.

The new complaint identifies 714 specific works in Exhibit A — including 'Wild Horses,' 'Sweet Caroline,' 'Bennie and the Jets,' 'Eye of the Tiger,' 'Bittersweet Symphony,' 'Viva La Vida,' and 'Radioactive' — and an additional 20,517 works in Exhibit B that Anthropic allegedly torrented to build a 'vast central library of written texts Anthropic would maintain forever.' The 40-fold expansion from Concord I to this filing repositions the case from a manageable licensing dispute into an existential legal threat.

The critical legal foundation comes from Judge William Alsup's July 2025 ruling in Bartz v. Anthropic. Alsup established that while AI training on legally acquired copyrighted content may qualify as fair use under existing doctrine, materials obtained through piracy are a standalone act of infringement that 'poisons the well' and does not receive fair use protection. The publishers are using this ruling to argue that even if Anthropic's training methodology could otherwise qualify as fair use, the piracy renders that defense unavailable.

Anthropic settled the Bartz v. Anthropic case for $1.5 billion, with impacted writers receiving approximately $3,000 per work for roughly 500,000 copyrighted works sourced from pirate libraries including LibGen and Pirate Library Mirror (PiLiMi). The settlement required Anthropic to destroy its pirated training data libraries within 30 days. The music publishers are applying the same $3,000-per-work rough valuation to their 20,517 works — but they are seeking the statutory maximum of $150,000 per work, not the settlement rate.

The complaint adds a count under the Digital Millennium Copyright Act for removal or alteration of Copyright Management Information, alleging Anthropic used algorithms that treated copyright notices as 'useless junk' and stripped them during the training process. Publishers also contend that despite Anthropic's claimed guardrails, current models including Claude 4.5 Sonnet, Claude 4.5 Haiku, and Claude 4.5 Opus can still be 'easily jailbroken' to reproduce copyrighted lyrics in full, and that the upcoming Claude 5 was trained on the same data.

The timing is not coincidental. The lawsuit was filed days after reports that Anthropic was seeking a new funding round at a $350 billion valuation — a 70-fold increase from its $5 billion valuation at the time of the Concord I filing in October 2023. The publishers explicitly noted Anthropic's market position in the complaint: 'While Anthropic misleadingly claims to be an AI safety and research company, its record of illegal torrenting makes clear that its multibillion-dollar business empire has been built on piracy.' Anthropic has not publicly responded to the expanded complaint.

The case tests whether the source of AI training data — legally acquired versus pirated — determines the entire copyright analysis, regardless of how the data is used afterward. If courts consistently apply the Alsup framework, every AI company that obtained training data through scraping pirate sites faces potential liability at scale. The case against Anthropic is the highest-stakes test of that principle to date.