Sony Music Publishing and Warner Chappell Sued Anthropic Over Lyrics, Seeking Up to $150,000 per Work

Sony Music Publishing and Warner Chappell Music filed a copyright infringement suit against Anthropic, chief executive Dario Amodei, and co-founder Benjamin Mann in the U.S. District Court for the Northern District of California on August 28, 2026. The publishers allege that Anthropic acquired and reproduced tens of thousands of musical compositions without authorization to train Claude and generated their lyrics as outputs, and they seek statutory damages of up to $150,000 per work willfully infringed plus up to $25,000 for each alleged removal of copyright management information. Anthropic said it disagrees with the publishers' claims and intends to defend itself robustly in court.

The Complaint Targets Acquisition, Not Training

The center of gravity in this complaint is how the training material was obtained rather than what Claude learned from it. The publishers describe a "brazen campaign of illegally torrenting, scraping and downloading copyrighted works on a massive scale," and characterize the conduct as one of the largest and most blatant ongoing thefts of intellectual property in history.

The alleged acquisition path is specific. According to the complaint, Benjamin Mann downloaded roughly 5 million pirated books from Library Genesis in June 2021 and obtained roughly 2 million more from Pirate Library Mirror in July 2022. Added to that are allegations of scraping lyrics from MusixMatch and LyricFind and of using the Common Crawl, The Pile, and Books3 datasets. The works at issue are the lyrics and sheet music carried inside those books.

Naming two founders as individual defendants follows from that structure. Once a complaint describes a named person downloading from a named site on a named date, rather than a corporate data-collection practice in the abstract, the dispute reaches the decisions behind the corporate entity directly.

The $1.5 Billion Settlement of September 2025 Set the Dividing Line

Anthropic settled a suit brought by book authors for $1.5 billion in September 2025, and in that case the court split the analysis between using copyrighted works for training and acquiring them through piracy. Training use was held lawful while acquisition through pirate sources was not. That split is why this complaint aims squarely at the acquisition path rather than at training.

The publishers turned that settlement back into an argument. The complaint states that Anthropic "clearly considers that to be just the cost of doing business," asserting that the settlement failed as a deterrent, which pairs with a damages theory built around statutory maximums.

The terrain hands the plaintiffs one favorable precedent and one unfavorable one at the same time. The unlawfulness of the acquisition route has already been recognized once, and so has the fair use character of training itself, in the very same case. The real front line is the narrow band between them: if acquisition was unlawful, how far does that reach into the training and the outputs built on top of it.

Why Lyrics Are Harder to Defend Than Books

Lyrics cases differ structurally from the book authors' case, and the difference lives in the outputs. The complaint alleges not only acquisition but verbatim reproduction of lyrics by Claude. Unlike a dispute confined to the transformativeness of training-data use, an established claim that source strings emerge intact destabilizes both transformativeness and market substitution, the two load-bearing factors in a fair use defense.

The physical scale of the work matters too. A novel runs to hundreds of thousands of characters, so whole-work reproduction is hard to posit, while a single lyric runs to a few hundred. Short works can be emitted in full, and full emission makes substantial similarity a simpler question. That asymmetry sits behind the music publishing industry suing earlier and more often than book publishers have.

Pleading copyright management information removal as a separate count belongs to the same logic. U.S. copyright law treats intentional stripping of authorship or rights information as a claim distinct from infringement, and this complaint prices it at up to $25,000 per instance. Training pipelines routinely drop attribution while normalizing text, which means an ordinary preprocessing step can become its own damages category.

Reading the "$150,000 per Work" Figure Correctly

The $150,000 per work figure is a statutory ceiling on what may be claimed, not a projected award. U.S. statutory damages cap at $150,000 per work only where willful infringement is found, drop to a much lower band when willfulness fails, and actual judgments or settlements typically land far from the ceiling. Multiplying tens of thousands of songs by the maximum produces headlines rather than a realistic coordinate for the case.

Two realistic coordinates already exist. Universal Music Publishing Group, Concord, and ABKCO have a suit in progress filed in January 2026 covering more than 20,000 works and seeking over $3 billion, and the book authors' case closed at $1.5 billion. The scale of this matter will likely form between those two numbers.

A third coordinate is the plaintiff lineup itself. With Sony Music Publishing and Warner Chappell joining, the publishing arms of all three major music companies are now litigating against Anthropic. That structure matters more than any individual damages figure, because it consolidates the seats on the other side of an eventual licensing table.

Three Things AI Teams Should Check Now

The practical takeaways for any team training or fine-tuning models are provenance records, preprocessing logs, and output filters. The complaint's core allegations attach at the acquisition stage and the output stage, so defensive evidence has to be manufactured at those same two points.

Provenance comes first. Which corpus was obtained when, from where, and under what license basis needs to exist as a record. The appearance of widely used public datasets such as Common Crawl, The Pile, and Books3 in this complaint restates that publicly available and permitted to use are not the same status.

Preprocessing logs come second. If a normalization step stripped authorship or rights notices, when and why it ran is the only defensive material available against a management-information claim. Output filters come third: independent of whether training data was lawful, whether a deployed service blocks verbatim reproduction of short works, and whether that blocking leaves a record, becomes a separate issue.

What Is Settled and What Is Not

What is settled is that the complaint was filed and what it demands, while the individual allegations, including the Library Genesis downloads, are the plaintiffs' assertions rather than judicial findings. Anthropic said it disagrees with the claims and intends to defend itself robustly in court, and fact-finding comes after proceedings begin.

Two questions stay open. The first is whether the fair use holding on training use from the book authors' case survives contact with short works and a verbatim-reproduction claim. The second is how far the naming of individual defendants holds up in actual litigation. Both are likely to be answered by the several concurrent suits against Anthropic together rather than by this one case.

Source: the copyright infringement complaint filed by Sony Music Publishing and Warner Chappell Music against Anthropic (filed August 28, 2026 in the U.S. District Court for the Northern District of California; defendants Anthropic, Dario Amodei and Benjamin Mann; tens of thousands of compositions at issue; statutory damages of up to $150,000 per work willfully infringed plus up to $25,000 per alleged removal of copyright management information; allegations of downloading roughly 5 million pirated books from Library Genesis in June 2021 and roughly 2 million from Pirate Library Mirror in July 2022, scraping lyrics from MusixMatch and LyricFind, use of Common Crawl, The Pile and Books3, and verbatim lyric reproduction in Claude outputs; the complaint's characterization of the conduct as one of the largest and most blatant ongoing thefts of intellectual property in history and its statement that Anthropic "clearly considers that to be just the cost of doing business"), Anthropic's public response, and prior case information (Universal Music Publishing Group, Concord and ABKCO suits of October 2023 and January 2026 covering more than 20,000 works and seeking over $3 billion, and the $1.5 billion settlement with book authors in September 2025) compiled by ASAP.