BMG Sues Anthropic Over Claude Training With 493 Alleged Copyright Violations

BMG filed a lawsuit against Anthropic in California federal court Tuesday, alleging the AI company used copyrighted song lyrics — from Bruno Mars, the Rolling Stones, Ariana Grande, and Justin Bieber among others — to train its Claude chatbot without authorization. The suit cites 493 examples of alleged infringement and seeks statutory damages of up to $150,000 per work, which could total more than $73 million if willful infringement is proven.

This is the second major music copyright action against Anthropic in three months. Universal Music Group, Concord, and ABKCO filed suit in January seeking $3 billion for alleged infringement of more than 20,000 songs. The BMG case is smaller in scale but adds a second plaintiff to a legal front Anthropic cannot easily dismiss.

THE NOVEMBER COMPLAINT DETAIL

BMG sent Anthropic a cease-and-desist letter in December 2025 demanding the company stop using its copyrighted compositions. Anthropic did not respond and refused to engage in licensing discussions, according to the complaint. The suit notes that Claude readily reproduces full lyrics to BMG-controlled songs including Uptown Funk, You Can't Always Get What You Want, and What A Wonderful World when asked — BMG argues this is evidence the model was trained on its copyrighted material.

The $150,000-per-work statutory damages ceiling matters because proving actual damages from training data use is difficult. Willful infringement — which BMG appears to be arguing given the December non-response and the LibGen allegations from the earlier UMG complaint — can trigger the higher end of the statutory range.

THE LIBGEN QUESTION

The expanded UMG complaint, filed in January and referenced in coverage of the BMG case, contains the most damaging specific allegation against Anthropic's founders. It alleges that Benjamin Mann, described as an Anthropic co-founder, personally downloaded approximately 5 million pirated books via BitTorrent from the library LibGen in June 2021 for use in AI training. CEO Dario Amodei personally discussed and authorized this illegal torrenting, the complaint alleged. The filing states that Amodei himself described LibGen as "sketchy" — and that Anthropic's own Archive Team had flagged the service as a blatant violation of copyright.

If those allegations are accurate, they undercut Anthropic's frequent defense that any infringing training data use was inadvertent or at arm's length. Anthropic did not respond to multiple requests for comment on this story.

THE LEGAL FRAMEWORK

Anthropic has argued that training AI models on copyrighted works constitutes fair use under U.S. law — that learning patterns and relationships from text is fundamentally different from copying the material itself. Courts have not ruled definitively on this question. The outcome of the pending cases could reshape how AI companies source training data.

One October 2025 ruling is already significant: U.S. District Judge Eumi Lee allowed music publishers to proceed with secondary liability claims against Anthropic, finding they had adequately argued the company could have known about and profited from user infringement. That ruling means discovery in the UMG case can explore what Anthropic knew about its training data sources — including potentially the LibGen downloads.

Anthropic reached a landmark $1.5 billion settlement with a group of authors in September 2025 over similar copyright claims. That settlement — a large number, but a fraction of what the music cases could cost if fully adjudicated — suggests Anthropic's legal team calculated that fighting all claims to judgment was more expensive than settling. The music cases are separate plaintiffs with different legal theories and damages claims, so the author settlement doesn't resolve them.

WHAT HAPPENS NEXT

The BMG case is BMG Rights Management LLC v. Anthropic PBC, U.S. District Court for the Northern District of California, No. 5:2026cv02334. The UMG lawsuit is Concord Music Group Inc v. Anthropic PBC in the same court, No. 5:24-cv-03811.

Anthropic will almost certainly argue these cases should be dismissed or stayed pending resolution of the fair use question at a higher court level. But the LibGen allegations — if they survive a motion to dismiss — could make that defense harder to sustain. A company whose co-founder allegedly personally torrented pirated books, authorized by a CEO who called the source "sketchy," has a harder time claiming innocent ingestion of training data.