Can AI companies train on my music without permission?

Where does the law actually stand?

The question — is training a music-generation model on copyrighted recordings fair use? — was put before two federal courts in June 2024. More than two years later, neither the District of Massachusetts (Suno) nor the Southern District of New York (Udio) has ruled on it. Sony filed a second Udio complaint on July 20, 2026 after the court refused to add 30,117 alleged recordings to the original case; that expands the dispute but decides nothing. Cases in other domains (text, books, images) have produced mixed, non-binding-on-music outcomes.

Where does the market stand?

While courts deliberate, the answer is being written in contracts: Warner settled with both Suno and Udio and licensed its catalog; UMG did the same with Udio; UMG and Roland published joint principles for music creation with AI. The direction is opt-in, paid, catalog-scale licensing. For an independent artist, that direction matters: the emerging norm treats training rights as something to be licensed, not taken — but until a court rules or Congress acts, an unlicensed company training on your tracks faces litigation risk, not settled illegality.

Your practical levers today are contractual (distributor and platform terms may address training) and collective (the opt-out and licensing frameworks being negotiated by rights bodies). No individual opt-out mechanism is enforceable by statute in the US as of our verification date.