Rights & policy

Sony and Warner's Anthropic lawsuit traces Claude's data pipeline

Thirty-five publishers accuse Anthropic and two founders of infringement across torrenting, scraping, training and model output. The complaint states no total damages figure.

AI-assisted

AI assisted with research organisation, structure and drafting; the editor checked the published text against the linked complaint, docket and reporting.

Light neo-expressionist painting of a music-processing machine and notation streams labelled INPUT?, LICENSE? and OUTPUT?

Sony Music Publishing, Warner Chappell Music and 33 affiliated publishers filed a copyright complaint in the Northern District of California on 28 August. The defendants are Anthropic PBC, chief executive Dario Amodei and co-founder Benjamin Mann.

The 48-page complaint attacks each stage of the data pipeline. It alleges that copyrighted compositions and lyrics entered through pirate book libraries, licensed lyric sites, third-party datasets and scanned physical books; that Anthropic copied the material during training; and that Claude later reproduced or adapted lyrics in its output. Anthropic told TechCrunch that it disputes the claims and intends to defend itself in court.

Four claims track the pipeline

Count I accuses all three defendants of direct infringement through BitTorrent. The publishers rely on conduct documented in the earlier Bartz authors case: at least five million books downloaded from Library Genesis in 2021 and at least two million from Pirate Library Mirror in 2022. They say those collections included lyrics and sheet music for hundreds of their compositions.

Count II asks the court to hold Amodei and Mann personally liable, alleging that they directed, approved and controlled the torrenting.

Count III is against Anthropic alone. It covers scraping lyrics, acquiring third-party datasets, scanning books, model training, verbatim and near-verbatim outputs, and derivative lyrics. The complaint says a non-exhaustive exhibit identifies tens of thousands of works.

Count IV alleges that Anthropic removed or altered copyright-management information while processing source text and generated output without the associated author and owner information.

The structure gives the publishers several routes to liability. A court could evaluate training separately from acquisition, distribution, output copying and metadata removal because the complaint pleads them as distinct acts.

The complaint states no multibillion-dollar demand

The publishers ask for up to $150,000 per work for willful copyright infringement, or actual damages and Anthropic's profits. They separately ask for up to $25,000 for each violation involving removed or altered copyright-management information.

The filing gives no total damages figure, and no money has been awarded. Multibillion-dollar headlines extrapolate from the phrase “tens of thousands” and the statutory ceiling. Any award would depend on which claims and works survive, what the evidence proves and which remedies the court permits.

The publishers frame AI training as a licensed market

Paragraph 12 says the publishers already grant licenses for authorised uses of their compositions in AI. They want AI developers to obtain permission on terms agreed with rightsholders. The complaint says Anthropic instead chose unpaid, allegedly pirated sources.

That position now spans the publishing arms of all three major music companies. Universal Music Publishing Group, Concord and ABKCO have two earlier cases against Anthropic, including a January 2026 action covering more than 20,000 works and seeking more than $3 billion. BMG followed in March, and Round Hill filed separate cases on 17 August.

None of the new Sony and Warner claims has been decided. Anthropic's formal response is the next substantive filing to watch on the court docket.

Original sources