Sony & Warner Sue Anthropic Over Alleged Pirated Music Used to Train Claude (2026)
Sony and Warner Sue Anthropic Over Alleged Pirated Training Data
Sony Music Publishing and Warner Chappell Music filed a federal copyright lawsuit against Anthropic on August 28–29, 2026, alleging the AI company illegally torrented, scraped, and downloaded copyrighted musical compositions to develop its Claude models. CEO Dario Amodei and co-founder Benjamin Mann are named as individual defendants alongside the company.
The complaint, filed in the U.S. District Court for the Northern District of California, describes what the publishers call “one of the largest and most blatant ongoing thefts of intellectual property in history.” Anthropic said it “disagree[s] with the publishers’ claims” and intends to defend itself in court.
Case at a Glance
Sony Music Publishing & Warner Chappell
Leading music publishers representing songwriting and composition rights.
Anthropic, Dario Amodei, Benjamin Mann
Company plus founders named on torrenting-related counts.
Up to $150K per work
Plus up to $25K per alleged DMCA copyright-management removal.
$1.5B (Bartz v. Anthropic)
Largest U.S. copyright class settlement—over pirated book downloads.
What the Complaint Alleges
The publishers claim Anthropic built Claude using copyrighted lyrics and compositions without licenses. The suit focuses on how training data was obtained—not only whether models were trained on copyrighted material.
| Alleged method | What publishers claim |
|---|---|
| BitTorrent / LibGen | Co-founder Benjamin Mann downloaded at least five million pirated books from Library Genesis in June 2021, including works containing lyrics and sheet music. |
| PiLiMi torrents | Anthropic employees torrented at least two million additional books from Pirate Library Mirror in July 2022. |
| Lyric scraping | Anthropic scraped licensed lyric sites including MusixMatch and LyricFind without authorization. |
| Book scanning | “Destructive scanning” of second-hand books to extract text. |
| Dataset pipelines | Use of Common Crawl, The Pile, and Books3 alongside the above sources. |
| Synthetic-data chain | Non-commercial models trained on pirated libraries allegedly fed commercial Claude releases through synthetic data and reinforcement feedback. |
Named compositions in the filing include 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—among tens of thousands of works the publishers say were copied.
Publishing Rights, Not Just Recorded Music
A common misconception is that this case is solely about audio files. The plaintiffs are music publishers—they control composition and lyric copyrights, distinct from record labels that own master recordings.
Sony Music Publishing and Warner Chappell administer rights for songwriters and composers. When AI models ingest lyrics from pirated books, scraped lyric databases, or torrented text collections, publishers argue that infringes their exclusive reproduction and distribution rights—even if no MP3 was downloaded.
Why the $1.5 Billion Book Settlement Matters
This music suit builds on precedent from Bartz v. Anthropic, the authors’ class action over Anthropic’s downloads from shadow libraries.
In June 2025, Judge William Alsup ruled that training large language models on lawfully obtained books could qualify as fair use because it is transformative. However, he held that downloading and retaining pirated copies from LibGen and PiLiMi was not fair use—it was “straightforward piracy.” That exposure led to a $1.5 billion settlement, approved in July 2026.
| Issue | Bartz ruling | Implication for music case |
|---|---|---|
| Training on licensed copies | Potentially fair use | Does not shield unlawful acquisition. |
| Torrenting from LibGen / PiLiMi | Infringement | Same conduct alleged in the music complaint. |
| Founder liability | Mann personally torrented millions of files | Amodei and Mann named again in the Sony–Warner suit. |
| Damages exposure | ~$3,100 per work in settlement | Music suit seeks up to $150,000 per work at statutory maximum. |
Music publishers are pressing the same weak point: even if AI training may be defensible in some contexts, piracy-sourced data is not.
The Broader Music Industry vs. Anthropic
Sony and Warner’s filing is the latest in a coordinated publisher campaign against Claude:
Concord / UMG I (2023)
~500 works; lyric reproduction and training claims. Summary-judgment fight on fair use ongoing.
Concord / UMG II (Jan 2026)
20,000+ works; ~$3B sought; torrenting and shadow-library allegations; founders named.
BMG (Mar 2026)
493 compositions; separate publisher suit in early discovery.
With Sony Music Publishing and Warner Chappell now suing, the publishing arms of all three major music companies are litigating against Anthropic.
Anthropic’s Response and Open Questions
Anthropic’s public statement is brief: it disagrees with the publishers’ claims and will defend itself robustly. The company has previously argued that commercial Claude models were not trained directly on LibGen or PiLiMi datasets, and that it later purchased legitimate copies of many books.
The Sony–Warner complaint challenges that framing—alleging that synthetic pipelines and non-commercial models still trace back to pirated sources. Discovery will determine which compositions appear in training corpora, whether outputs reproduce lyrics, and whether founders bear personal liability.
Nothing in the filing proves infringement at trial. Allegations remain allegations until a court rules.
What This Means for AI Training Data
From free scrape to paid input
Publishers with legal catalogs are positioned to license data—or sue. Training data is shifting from a public commons to a line-item cost.
Acquisition matters as much as use
The Bartz split suggests courts may accept transformative training while rejecting pirated acquisition. AI labs are auditing data provenance accordingly.
Multi-front exposure
Anthropic faces parallel music cases, a prior billion-dollar book settlement, and ongoing fair-use battles—raising the stakes for every major model builder.
Conclusion
The Sony Music Publishing and Warner Chappell lawsuit marks a new front in the fight over AI and copyright. It alleges the same torrenting conduct that cost Anthropic $1.5 billion in the book authors’ case—now applied to tens of thousands of musical compositions tied to Claude.
For the industry, the message is clear: who owns the rights to songs and lyrics intends to be paid—or heard in court. How Anthropic and other AI companies source training data will shape the economics of the next generation of models.
