Universal Music Publishing Group, Concord Music Group, and ABKCO sue Anthropic for allegedly violating their copyrights by using song lyrics to train AI models
Murray Stassen / Music Business Worldwide :
Context & Ripple Effects
The publishers had already pressed music platforms to stop third-party developers from scraping copyrighted melodies and lyrics for AI training, as reflected in UMG's requests to platforms to curb music-data scraping. This lawsuit moves that pressure from distribution intermediaries to a model developer.
The dispute became a durable test of the training-data question: Anthropic later defeated an attempt to immediately halt its use of the publishers' lyrics, but the publishers subsequently expanded their campaign with a second Anthropic lawsuit involving more than 20,000 songs.
First-order effects
- Anthropic must defend its lyric-training practices against claims from UMPG, Concord and ABKCO, while the publishers seek to establish that their catalogs require authorization for this use.
- The case puts the commercial value and provenance of lyric datasets directly at issue for the named publishers and Anthropic.
Second-order effects
- Other AI developers using music-related text face greater pressure to document training sources, negotiate permissions, or contest similar claims.
- Music rightsholders gain another litigation lever alongside the licensing route later discussed for AI music startups, including talks to license major-label music to Udio and Suno.
Third-order effects
- If courts or settlements consistently favor rightsholders, copyrighted works may shift from broadly available training inputs toward licensed, auditable AI inputs.
- The music sector could become an early template for how generative-AI companies balance model development against catalog licensing and copyright risk.
The trend: This is part of the broader shift from unlicensed content collection toward negotiated or litigated access to high-value training data.