On August 28, 2026, Sony Music Publishing and Warner Chappell Music filed suit against Anthropic in the U.S. District Court for the Northern District of California, accusing the AI company of what the complaint calls one of the largest and most blatant ongoing thefts of intellectual property in history. The publishers allege Anthropic ran a brazen campaign of illegally torrenting, scraping, and downloading copyrighted song lyrics to train its Claude models. With this filing, the publishing arms of all three major music companies are now litigating against Anthropic at once, turning a series of separate complaints into a coordinated legal front.
What the Complaint Alleges
The heart of the claim is straightforward, even if the law around it is not. The publishers say Anthropic trained Claude on tens of thousands of copyrighted musical compositions without a license, and that it obtained many of them from pirate sources. The complaint names repositories including Library Genesis and a Pirate Library Mirror, alongside lyric databases such as MusixMatch and LyricFind, as sources the publishers say were used improperly. A second strand of the claim is about copyright management information, the metadata that identifies who owns a work; the publishers allege Anthropic stripped that information out, which is a separate violation under U.S. copyright law.
Those two strands map onto two different money claims. For the underlying infringement, the publishers seek statutory damages of up to $150,000 for each work infringed willfully, the top of the statutory range. For the removal of copyright management information, they seek $25,000 per instance. Multiplied across tens of thousands of compositions, either figure implies potential exposure in the billions, which is why the complaint is described as seeking multi-billion-dollar damages and demands a jury trial.
A widening front: music-publisher suits against Anthropic
The lyric-training dispute has escalated from a single 2023 complaint to litigation from all three major publishers.
- Oct 2023UMPG, Concord, ABKCO
The first publisher suit over lyrics used in training.
- Jan 2026Follow-on complaint
A second suit reported to seek more than $3 billion.
- Mar 2026BMG
Another major publisher joins the litigation.
- Aug 17, 2026Round Hill Music
A further publisher files against Anthropic.
- Aug 28, 2026Sony Music Publishing + Warner Chappell
The two file suit, so the publishing arms of all three major music companies are now litigating against Anthropic.
The Bigger Picture: A Coordinated Front
This suit does not stand alone. It is the culmination of a widening wave. The first publisher case, brought by Universal Music Publishing Group, Concord, and ABKCO, dates to October 2023. A follow-on complaint in January 2026 was reported to seek more than $3 billion. BMG joined in March 2026, and Round Hill Music filed on August 17, 2026. Sony Music Publishing and Warner Chappell's August 28 filing completes the set, bringing the publishing operations of all three majors into active litigation against the same defendant.
A series of separate complaints has become a coordinated legal front. That changes the negotiating dynamics as much as the legal ones.
Analysis of the music-publisher litigation against Anthropic
That coordination matters beyond the courtroom. Copyright cases of this size rarely run all the way to a verdict; they tend to resolve through licensing settlements. When every major publisher is litigating at once, the industry is effectively negotiating the terms on which AI companies will, or will not, be allowed to train on music going forward. The lawsuits are the leverage; a licensing framework is the likely destination. Reporting notes that the publishers also drew on evidence unearthed in an earlier case, the Bartz litigation, which suggests the plaintiffs are building on a shared factual record rather than starting from scratch.
Why This Is Legally Unsettled
It would be a mistake to treat the allegations as a settled verdict. They are claims, and Anthropic has not been found liable. The central legal question, whether training an AI model on copyrighted text is fair use, remains genuinely unresolved in U.S. law, with different judges reaching different conclusions on related questions. What tilts this particular case is the alleged sourcing. There is a meaningful legal difference between training on works a company lawfully accessed and training on works obtained from pirate repositories. Courts have signaled that how the training data was acquired can matter a great deal, even where the ultimate fair-use question is contested. The publishers' emphasis on torrenting and pirate mirrors is a deliberate move onto that firmer ground.
The copyright-management-information claim is a second, quieter pressure point. Even in a world where some training use is eventually deemed fair, stripping ownership metadata is its own alleged act with its own statutory penalty, and it is harder to defend as incidental. Together, the two theories give the publishers multiple ways to win something even if the headline fair-use question breaks in Anthropic's favor.
What It Means for AI Builders and Buyers
For the broader industry, this case is another data point in a trend that is quietly reshaping how frontier models get built: training data is becoming a licensed, negotiated input rather than a free one. The era in which labs could scrape broadly and litigate later is closing, and the direction of travel is toward paid arrangements with rights holders, at least for high-value, clearly-owned content like commercial music. That has cost and competitive implications. Labs that secure licenses gain legal certainty; labs that do not carry a growing liability on their balance sheets.
For companies that buy and deploy AI rather than build it, the practical exposure is indirect but real. Legal uncertainty around a provider's training data is a form of vendor risk: an adverse ruling or a large settlement can affect a model's economics, its availability, or the terms under which it is offered. It is one more reason not to hard-wire a critical workflow to a single model. Working across models from multiple providers, the approach behind model-agnostic platforms like Metir AI, spreads that exposure rather than concentrating it in one lab's legal outcomes.
None of this resolves the underlying tension, which is a real one. Creators and publishers have a legitimate claim that their work has value and should not be used to build commercial products for free. AI developers have a legitimate interest in learning from the broad corpus of human culture. The courts are now being asked to draw that line for music specifically, and with all three major publishers at the table, the line they draw will shape the terms for the entire industry.
Sources:
- Now Sony Music Publishing and Warner Chappell sue Anthropic in multi-billion dollar lawsuit | Music Business Worldwide
- Sony Music, Warner sue Anthropic, alleging a "brazen campaign" of intellectual property theft | TechCrunch
- Sony, Warner sue Anthropic, alleging "blatant theft" of intellectual property | Axios
- Sony and Warner sue Anthropic for 'blatant violation' of copyright law | Engadget
- Music publishers use evidence unearthed in Bartz to hit Anthropic with $3B suit | California Courts Newsroom
Image credits
Header image: an audio engineer mixing at a recording console at Jackpot! Recording Studio in Portland, Oregon, by VACANT FEVER via Wikimedia Commons, licensed under CC BY-SA 2.0. In-body photograph of the Phillip Burton Federal Building and United States Courthouse in San Francisco, by Marincyclist via Wikimedia Commons, licensed under CC BY-SA 4.0.