How Musicians Can Get Paid for Training AI
TL;DR
IEEE Spectrum outlines a model where musicians are paid not only once for AI training, but also when their work measurably influences generated music later. Sureel, now acquired by Warner Music Group, is working with Swedish rights agency STIM on media labels, usage rules, tracking and training-data licensing fees. SoundVerse rejects one-time buyouts and argues for ongoing artist participation: training data that contributes more to an output should earn more.
Nauti's Take
Attribution sounds like the cleanest answer, but it is not yet a fair market. Whoever measures which song really influenced a model also decides who gets paid.
That could strengthen artists, or simply create another opaque royalty machine. Smaller licensed models, musician collectives and auditable rules look more promising because credits and money flows can stay visible.
Briefingshow
The issue is not only copyright, but the economic logic of creative work: people who create value should share in downstream revenue. If AI models absorb music and turn it into new products, a one-off payment may not be enough. Without transparent rules, power and margin shift toward labels, platforms and AI companies.
Sources