The doctrine
UMG, Sony Music, and Warner Records have sued the leading generative-music startups (Suno and Udio) over alleged unlicensed use of master sound recordings to train models. Major label settlements with Udio reshaped the music-AI licensing landscape in 2025.
The music-AI cases are factually distinctive. Both Suno and Udio admitted in early filings that their training corpora included copyrighted recordings, isolating fair use as the dispositive defense. Plaintiffs supported their complaints with side-by-side spectrogram exhibits showing generated outputs that closely matched identifiable hits.
Outside the master-recording cases, music publishers (Concord Music, Universal Music Publishing) have sued Anthropic and OpenAI over reproduction of song lyrics. The lyric cases sit at the intersection of training-data and output-infringement theories: lyrics are short enough that even partial regurgitation can capture protected expression in full.
The licensing-deal model reportedly used in the Udio settlements — combining cash payments and equity with prospective licensing — is now the labels' template for negotiating with surviving generative-music companies.
Leading cases
Lead RIAA-coordinated music-AI suits; Warner and (for Udio) UMG settled October 2025.
Lyrics-publisher action against Anthropic; §1202 narrowed at pleadings.
European collecting-society action; first ruling that LLM lyrical output infringes.
Key holdings
- Spectrogram evidence works. Side-by-side spectrogram exhibits have become standard exhibits in music-AI cases.
- Settlements include licensing. Major-label settlements increasingly bundle damages with prospective licensing and equity.
- Lyrics cases are output cases. Short-form lyrics make output infringement easier to plead than for long-form text.