How Musicians Can Get Paid for Training AI
TL;DR
Music already has a dense payment system: sales, streams, radio play, covers, karaoke and sampling can trigger royalties depending on how a work is used. AI training does not fit neatly into that system yet. The hard question is what counts as use. Was a song used once during training, or does its creative influence keep working every time the model generates new music? Sureel and SoundVerse are trying to make that influence traceable and pay rightsholders when generated tracks rely on licensed musical references.
Nauti's Take
The fair deal has to look more like music economics than software economics. A one-time training fee would be too cheap for many catalogs because the value keeps working inside the model afterward.
A better path is licensing that makes references, permitted uses and payouts visible. The real test is not the pitch deck, it is accounting, audits and disputes.
Briefingshow
If AI music systems learn from existing catalogs, a new market opens next to streaming and sync licensing. The core question goes beyond copyright: who decides whether a work can be used for training or as a reference, and who gets paid later? For smaller musicians, that infrastructure matters because opt-out forms do not create income.
Sources