Licensing music-adjacent audio for AI

Music-adjacent audio is everything that sings, plays, or performs without being a finished commercial song: isolated vocal stems, session and rehearsal recordings, live performance tapes, demos, and designed sound. Legally it is thicker than speech for a stackable reason. A voice memo implicates one copyright and one person. A recorded performance of a song implicates at least two copyrights — the sound recording and the underlying musical composition, which US law treats as separate works with typically separate owners — plus the performers’ voice and publicity rights, plus, where the players were union, collective-bargaining obligations that follow the recording into new uses. The US Copyright Office’s guidance is built around that recording/composition split: licensing the master without the composition clears half the work. This is why "we have the audio file and the owner signed" is a complete answer for speech and an incomplete one for anything musical.

Why it matters to a buyer

The expensive mistake is treating a licence to the recording as a licence to the music. The composition travels inside every stem: a vocal stem embodies the melody and lyrics, and the publisher’s rights in them are not cleared by the master owner’s signature — the publishers’ lyrics suits against Anthropic, separate from and unresolved by the authors’ book settlement, are the live demonstration. The 2025-26 market has also now priced unlicensed music training. The major labels sued Suno and Udio in June 2024, seeking statutory damages up to $150,000 per infringed recording; the endgame so far has been conversion, not verdicts. Universal settled with Udio in October 2025, pairing a compensatory settlement with recorded-music and publishing licences for a new platform whose outputs stay inside the service. Warner settled with Suno in November 2025 — licensed models with artist opt-in on voice and likeness, launching in 2026 — and separately resolved its Udio claims the same month. Sony settled with neither and is still litigating, and GEMA’s Munich case against Suno awaits judgment. Two cautions from that history. First, settle-and-license is the retrofit price of having trained first; licensing first is cheaper. Second, a label licence does not necessarily end the matter: in June 2026 the American Federation of Musicians sued Universal and Warner, alleging the Suno and Udio deals put members’ recorded work to a new use requiring compensation under the union’s agreements. A catalog licence that ignores the people who played on the recordings can leave the performer layer unpaid — and litigating.

Why it matters to a data owner

The stack that burdens buyers is the supplier’s pricing power, if the paperwork is complete. The cleanest supply in the market is material where the layers collapse: a creator who wrote the composition, performed it, and owns the recording holds both copyrights in one pair of hands and can grant training rights nobody else can contest. For anything involving other people, document by layer: who owns the recording; who owns the composition; who sang and played, with separate voice consent for identifiable singers, since state statutes such as Tennessee’s reach voice directly; and whether any session ran under a union contract. The SAG-AFTRA Sound Recordings Code, ratified in April 2024 with a dedicated AI appendix, requires clear and conspicuous written consent, in a separate writing and per project, before release of a covered sound recording using a digital voice replica — and notably does not itself cover consent to train on a voice, which means training consent has to live in your own paperwork rather than be assumed from union coverage. Suppliers who arrive with the full grid — both copyrights, performer consent, union status stated — are selling exactly what the settlements taught buyers to demand.

Current legal status

US copyright treats the sound recording and the musical work as separate copyrights; the Copyright Office’s Circular 56A sets out the distinction. On litigation and licensing, verified as of July 2026: the major labels filed against Suno (District of Massachusetts) and Udio (Southern District of New York) on June 24, 2024, seeking statutory damages up to $150,000 per work. Universal settled with Udio on October 29, 2025, combining a compensatory settlement with recorded-music and publishing licences for a licensed platform. Warner settled with Suno in November 2025 — licensed models with artist opt-in over voice and likeness, planned for 2026, with current models to be deprecated — and separately resolved its case against Udio the same month; Udio has since added licensing agreements with Merlin and Kobalt, among others. Sony Music remains in litigation against both companies. The publishers’ lyrics cases against Anthropic remain live and are covered separately on this site. On the performer layer: the American Federation of Musicians sued UMG and Warner Music Group on June 5, 2026 in the Southern District of New York, alleging the Suno and Udio licences triggered the new-use compensation provisions of its collective agreements; the case is in its early stages. SAG-AFTRA’s 2024 Sound Recordings Code with the major labels, ratified April 30, 2024, added an AI appendix requiring separate, per-project written consent before release of covered recordings using digital voice replicas, with minimum rates for replica use. In Germany, GEMA sued Suno in the Munich Regional Court in January 2025 (case 42 O 763/25); after a March 9, 2026 hearing, judgment is scheduled for July 31, 2026. Nothing in the settlements creates precedent; they create prices.

What fiund does about it

fiund lists music-adjacent audio only with the stack resolved: recording and composition rights addressed in the licence, separate consent for identifiable performers, union status documented, provenance on request. That is the difference between an audio file and a licensable asset.

Sources

← All rights & provenance guides