The AI policy: what is welcome, and what is refused

Yes. AI-assisted and fully AI-generated music is welcome on Run, on every plan, and it goes to the same stores as everything else.

Disclosure is the condition. For every release you say how much AI is in the artwork, in the recording and in the composition, and at least one credited person has to be a human. That is the whole bargain.

What Run refuses is not a level on the disclosure scale: every level is accepted and distributed. What is refused is conduct, and it comes down to two things. Using a real person’s voice or identity without their permission, and producing material to collect plays rather than listeners.

What is welcome

• AI-assisted composition and songwriting. Prompting, drafting, rewriting, arranging.
• AI production tools on your own work. Mixing, mastering, stem separation, noise removal, restoration, synth and sound design.
• AI elements combined with human performance. A generated bed under your own vocal, a generated instrument in an otherwise played arrangement.
• Original synthetic voices. A voice that is not built to sound like a specific real person is yours to release.
• Original AI artwork. The cover follows the same artwork rules as any other. See artwork requirements.
• Fully AI-generated recordings and compositions, disclosed as such, with a human credited on the track.

The rule: disclose it accurately

Three questions travel with every release: one about the artwork, and one each about the recording and the composition on every track. Each uses the same four-step scale, from no AI to fully AI. Every credited person also carries their own answer about their own contribution.

Answering honestly costs you nothing. Nothing on Run charges more, delivers slower or reaches fewer stores because a disclosure says AI. AI disclosure, explained covers what each level means and where you answer it.

What is refused

None of these have a disclosure that makes them acceptable.

• AI voices imitating a real artist. Passing your record off as someone else’s performance is impersonation, whatever tool made it.
• Deepfake and cloned vocals of an identifiable person. Same reason, and it is the complaint stores and rights holders act on hardest.
• Soundalikes intended to pass as someone. A recording made so a listener, a playlist or an algorithm mistakes it for another artist’s work.
• Any use of an identifiable person’s voice, name or likeness without their explicit written consent and a signed agreement. Consent is a document, not an assumption.
• Mass-produced material made to inflate streams. Bulk near-identical tracks, functional filler and catalogs generated to farm plays take money out of a pool every other artist is paid from.
• Output that reproduces copyrighted material. A model trained to reproduce a specific protected work, and any output that does so, is the same infringement it would be by hand.

Under-disclosure and mis-disclosure

Declaring less AI than is really in the work is a breach of the agreement you accept, and it gets the release held rather than shipped. Some of it is caught automatically:

• A track where every credited person is marked fully AI does not submit. Someone chose the prompt and decided it was finished.
• A credit marked as AI-involved while both of that track’s disclosures say none is flagged for a person to read.
• Titles in the “AI cover” or “AI voice” format fail outright: that is the marketing format of voice-clone releases. The vaguer “AI version” and “AI remix” wording is flagged for a reviewer instead.
• “In the style of” in a title fails, with or without AI, because it reads as a soundalike.
• A track disclosing AI-generated audio with no voice or likeness consent document attached goes to a reviewer. There is no field anywhere that says “this is a real person’s voice”, so a person checks rather than software guessing.

A flag is not a refusal. It goes to the person reviewing your release with everything else the checks found. See what our review checks.

If your track uses a real voice with permission

Attach the consent. In the release form, the AI voice and likeness consent document goes on the track it belongs to, alongside the other clearance documents. Without it, a reviewer has no way to tell your licensed feature from a clone. Covers, remixes and samples explains how documents are attached.

Where your answers go

Your disclosures are delivered with the release. Each store then applies its own AI policy to them, and those policies are still moving: stores have been publishing, revising and enforcing AI rules of their own, and we track what they publish and keep these rules in step with it. Passing our checks is never a promise that a store will accept a release or keep it up.

Two consequences worth knowing before you submit. A track whose recording is disclosed as fully AI cannot be opted into Content ID, because fingerprint networks require exclusively owned original recordings; see stores and Content ID. And changing an AI disclosure on a release that is already out sends it back through a fresh review in our distribution pipeline, which can affect availability while that review runs.

Tell us anyway if an answer was wrong. An honest correction is always better than a wrong disclosure sitting on a live release.