Voice cloning consent and ethics — permission, policy, disclosure

A voice clone takes a minute of audio and about as long to make. That asymmetry — cheap to create, expensive to misuse — is the whole reason this article exists. None of what follows is complicated, and most of it is one decision made once, at the start, about whose voice goes into the form.

Two voices you may clone, and no third

Your own, or one you have written permission to use. That is the entire list.

Everything else is a variation on the same refusal. A celebrity, a streamer, a politician, a colleague who has not been asked, an ex, a voice lifted from a podcast, a voice lifted from a video call: cloning a real, identifiable person without their consent is banned outright by the acceptable use policy, alongside impersonation and deepfakes generally. Enforcement happens at the account level, which is the part people underestimate — it is not the clip that goes, it is the account and the credits in it.

The platform is for adult fiction. Your narrator, your characters, your own ad read, a collaborator who said yes in writing: those are the jobs voice cloning exists for.

Get the permission in writing, and make it specific

"They said it was fine" is not a record. A permission note does not need a lawyer, it needs to be dated, specific and kept. Six lines is enough:

I, Dana Ruiz, give Mara Kessler permission to create an AI voice clone from the recordings I provided on 2 September 2026, and to use that voice in the audio drama series Nightline, published on her own channels and podcast feeds.

This permission does not cover advertising, political content, or making me appear to say things as myself.

I can withdraw it at any time by email, and the clone will be deleted within seven days of my asking.

Signed, Dana Ruiz — 2 September 2026

Four things make it useful: what the voice may be used for, what it may not be used for, how consent is withdrawn, and a deletion commitment you can actually honor. If you cannot picture yourself deleting the clone on request, do not make it.

One more clause worth naming out loud in the note: the clone must not be used to make the person appear to say things as themselves. A voice performing a fictional character is a different act from a voice making a statement, and a permission that does not separate them will eventually be tested.

Where the clone actually lives

Knowing this makes the deletion promise easy to keep.

A clone is private to your account. It appears in your voice library next to the roster voices, Aria and Orion, and nobody else can list it, hear it or render with it — it is scoped to the account that made it, not shared into a pool other users draw from.

Deleting it is one button in that library, or one DELETE /v1/voices/{id} call from the speech API. The clone is removed from the speech engine as well as from your library, so it stops being renderable, not just stops being listed. Audio you already rendered is unaffected — those files exist and are yours to keep or pull down, which is why the permission note should say what happens to published episodes, not only to the clone.

Cloning itself is free. That matters for consent, because it means "delete it and re-clone later if they change their mind" costs nothing but the sample.

What a responsible disclosure sounds like

Disclosure is not a confession, and it does not have to be an announcement. Match the weight of it to the weight of the deception risk — a narrator reading your own words in your own cloned voice is a low-risk case; a voice that listeners will assume belongs to someone else is a high-risk one.

Three versions that work, in ascending order of visibility:

Narration is an AI voice clone of the author's own voice.

Voice of Dana in this series is an AI clone of Dana Ruiz, made and used with her permission.

Everything you are about to hear is fiction, performed by synthetic voices. No real person said any of it.

Put it where the audience already looks — the episode description, the end card, the pinned comment, the credits. A line in a footer nobody reads is a disclosure in name only. And if the piece is likely to be clipped and reshared, get the disclosure into the audio, because a description does not survive a repost.

The cases worth refusing

A short list, learned the boring way:

• A voice you found. Public audio is not public permission. A podcast feed, a Twitch VOD and a voicemail are all someone's voice.
• A "just for us" clone of a friend. Private intent does not survive a leaked file, and the account-level rule does not care about intent.
• A voice of someone who cannot consent. No consent, no clone. There is no workaround worth the account.
• Anything that puts words in a real person's mouth as themselves. Even with permission for other uses, this is the one that turns into a news story.

When a project cannot clear consent, the roster is the answer rather than the obstacle. Aria and Orion are the platform's own voices, read 13 languages, and carry no permission chain to maintain — and a character with a roster voice is a character nobody can claim to be.

Build the habit into the workflow

Name clones after the consent, not the person — "Dana, Nightline, consent 2026-09" tells you in six months exactly what you are allowed to do with it. Keep the permission notes in the same folder as the project files. Delete clones when a project ends rather than letting a library accumulate voices whose paperwork you can no longer find. The voice cloning guide covers the recording and rendering side once the consent question is settled.

FAQ

Can I clone a celebrity voice for parody or fan work?

No. Real, identifiable people without consent are banned regardless of intent, including parody, tribute and fan projects, and enforcement is at the account level. If a project needs a distinctive voice, cast a roster voice or clone your own and perform the part.

Are my cloned voices private, and can I delete them?

Yes to both. A clone is scoped to your account, is not shared or pooled, and can be deleted from your voice library or with a single API call, which removes it from the speech engine rather than just hiding it. Audio you already rendered stays in your hands.

What happens if a request is blocked by policy?

Blocked requests return a content_blocked error and are never charged, so a refusal does not cost credits. Repeated attempts are a different matter — enforcement is at the account level, not the request level.

Do I have to tell my audience a voice is AI?

Platform rules and local law vary, so check the places you publish, but the practical standard is simple. If a listener could reasonably believe a specific real person spoke those words, say so, in the audio as well as the description.

Settled on whose voice it is? Open the Voice studio and start with the sample, not the script.