Policy & Safety
The training-data wars end not with a verdict but a market
Collective licensing bodies now sell blanket training rights for text, images and music, turning years of litigation into a functioning royalty system.
By Sofia Lindgren, AI Policy Correspondent — LONDON
LONDON — The great copyright confrontation over AI training data is ending the way music's did: with collecting societies. Blanket licensing bodies representing publishers, image libraries and record labels now sell standardised training rights, and every major AI lab has signed at least one.
The economics echo radio royalties — per-use micro-payments pooled and distributed by audited share of training corpora — and follow court rulings in several jurisdictions that left both sides preferring a market to further coin-flips.
Individual creators remain split: payments are real but modest, and opt-out rights — the movement's original demand — survive mostly for future works. The unresolved frontier is synthetic output that competes with the licensed inputs that trained it.
Enable JavaScript to read the full story on Neural Daily News.