Music publishers sue Anthropic over song lyrics

Universal Music Group, Concord Music Group and ABKCO sued Anthropic in federal court in Tennessee, alleging the company’s Claude models had been trained on and could reproduce copyrighted song lyrics without a licence. The complaint said Claude had been trained on lyrics from at least 500 songs by artists including Katy Perry, the Rolling Stones and Beyoncé, and cited an example in which the model produced what the publishers described as a near-identical copy of the lyrics to Katy Perry’s “Roar.” The publishers sought statutory damages of up to $150,000 per work infringed, alongside injunctive relief.

The suit distinguished itself from the wave of author and news-publisher litigation against AI companies that year by targeting lyrics specifically — a category with its own licensing market and its own history of aggressive enforcement, since music publishers had long pursued unauthorised lyric reproduction on ordinary lyrics websites. The case was subsequently transferred out of Tennessee to the Northern District of California, consolidating the dispute closer to where Anthropic and most of the other AI copyright litigation was based.

Anthropic did not dispute that some copyrighted lyrics may have appeared somewhere in its training data, but argued that reproduction by the model was rare, that its usage policies prohibited using Claude to infringe copyright, and that it had built guardrails to prevent verbatim output. A federal judge later ordered Anthropic to maintain and extend certain output-filtering safeguards while the case proceeded, but in March 2025 rejected the publishers’ bid for a broader injunction, finding they had not shown the existing licensing market for song lyrics was being undercut by Anthropic’s practices.

The case remained one of several parallel copyright actions against Anthropic — alongside the separate authors’ class action that would produce a landmark settlement in 2025 — testing how far fair-use defences for AI training would extend across different categories of copyrighted content.