Resources
Podcast guest release forms for AI training
A standard appearance release rarely covers model training. Here is what to add, and why forward consent beats chasing it later.
Published 2026-07-22 · 6 min read
Key takeaways
- A publishing release does not cover AI training; the permission must be named.
- Forward consent at signing is far easier than chasing former guests later.
- Capture voice and likeness consent separately from the content licence.
- Let the release allow licensing to a third-party model builder.
- Keep an auditable record of who agreed to what, and when.
Most podcasters already use a guest release. It covers publishing the episode, promoting it, keeping it online. Useful. But it was written for distribution, not for AI training.
If you want to license episodes as training data, the release has to reach further. It has to name the use, cover commercial models, and allow the material to be cleared for a buyer. This is what changes.
Why a standard release falls short
A typical release grants the show the right to record, edit, and distribute a guest’s contribution. That is about publishing. Training a model is a different use, and an older form usually does not mention it.
Courts and lawmakers increasingly treat a person’s voice as protected. Tennessee’s ELVIS Act extended that state’s right of publicity to cover voice, aimed squarely at AI voice clones. A proposed federal bill, the NO FAKES Act, would create a nationwide right over unauthorised digital replicas of voice and likeness. The direction of travel is clear. Consent to publish is not consent to train.
Forward consent versus chasing it later
There are two ways to get AI-training consent. Ask at the time. Or go back later and ask again.
Forward consent is far easier. You add the language to the release the guest signs before recording. Everyone is present, the terms are fresh, and the file is clean from the start.
Retroactive consent means finding a guest from years ago, explaining a use they never anticipated, and hoping they agree. Some will. Many will not reply. Every guest you cannot reach is an episode you cannot license. Build consent in at the front, and you avoid the chase.
What AI training adds to the ask
The core addition is a clear, specific permission to use the recording to train, develop, and evaluate AI or machine-learning models. Not implied. Named.
Alongside it sits voice and likeness consent. Where a guest is identifiable, their voice is a protected attribute in its own right. The content licence and the voice or likeness consent are two permissions, and both should be captured.
The release should also allow the material to be licensed to a third party — the model builder — because the guest is not signing with the buyer directly. And it should confirm the use may be commercial, so no one later argues they only agreed to research.
Where fiund fits
You do not have to draft all of this alone. On fiund, consent is captured and recorded at the source, per asset. Owners keep ownership of their recordings and approve which buyers may license them.
Because the platform handles voice and likeness consent separately where a person is identifiable, and keeps provenance a buyer can inspect, the clearance work that would otherwise stall a deal is done up front. Your job is to make sure the people in the room agreed. The platform carries that agreement through to the licence.
Clauses an AI-ready guest release should cover
- A named, specific right to use the recording to train, develop, and evaluate AI or machine-learning models.
- Voice and likeness consent for identifiable guests, separate from the content grant.
- Permission to license the material to third parties, including model builders.
- Confirmation that use may be commercial, not research-only.
- The territory and duration the consent covers.
- Whether the guest may withdraw, and what happens to models already trained.
- A record of who consented, to what, and when — kept with the file.
Sources
Frequently asked questions
Can I reuse my existing guest release for AI licensing?
Only if it already names AI training, allows commercial use, and permits licensing to third parties. Most older releases do none of these, so they need an update or a short addendum.
What if a past guest will not respond to a re-consent request?
Then you cannot license that episode. This is why forward consent matters. An unreachable guest is an asset you have to leave out.
Do I need voice consent as well as a content licence?
For identifiable guests, yes. The voice is a protected attribute on its own. Capture that consent separately so the record is clear.
Related resources
Want data that clears this in diligence?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief