Someone asks you to make a voiceover "in my voice" so they never have to record again. Or your client wants their founder's voice on a hundred support calls. Cloning a voice takes minutes. Deciding whether you should, and proving you were allowed to, takes a bit of paper. This post is that paper: what ElevenLabs' own policy and terms say, a consent form you can adapt, and a checklist to run before you press clone.
This is not legal advice. Voice, likeness and privacy law vary by country, and I haven't researched any of them here. I downloaded three ElevenLabs pages on 2026-10-09 and read their text (the use policy, the terms of use and the voices documentation), and I paraphrase them below; the wording can change, so open the pages and read the exact text before you rely on it. I did not clone any voice in this session.
What does ElevenLabs actually require?
Three documents matter. Each said the following when I read it.
Prohibited use policy (page marked last updated 17 August 2026). Under its voice cloning and impersonation section, it prohibits replicating another person's voice without their consent or legal right, including to act on their behalf, and using a replicated voice to harass or harm that person. It also prohibits replicating a voice in a manner intended to deceive others about whether it was generated by AI. Under political content, it prohibits impersonating political candidates or elected officials "regardless of whether authorization was obtained". Under deception and fraud, it prohibits scams and evading product guardrails such as voice verification. Under other requirements, organisations using the services, including to power AI agents, must clearly and prominently tell users they are speaking with AI. Violations can lead to removed material or suspended access.
Terms of use (page marked last updated 31 March 2026). On voice models, it says you may be asked to upload recordings of your own voice or one you're authorized to share, and may upload only that. You grant ElevenLabs a broad licence to your content and to your voice models, described in the terms as perpetual and irrevocable, covering use to provide and improve the services and to develop new ones; the terms add that the company will not commercialise your voice on a standalone basis without your permission. That matters for the consent form below: a voice owner can withdraw permission from you, but a withdrawal clause in your paper cannot undo a licence you accepted from the platform, so tell the voice owner this before they sign. You warrant that you have the rights needed and that your content and voice models won't violate anyone's rights, and you indemnify ElevenLabs for claims from your content or use. You can request deletion of voice models made from your recordings in your account, and the Privacy Policy's voice processing notice covers retention.
Product documentation (voices overview, read the same day). It describes instant voice cloning from short audio and professional voice cloning that needs longer training audio and a Creator plan or above. It states that professional clones use voice-captcha technology to verify they're created from your own voice samples, and it describes no such step for instant clones. Professional clones can be shared publicly in the Voice Library, instant clones cannot. Pages for the specific instant and professional cloning guides returned "page not found" when I tried them, so I can't tell you more about those flows.
Two things follow. The policy puts the burden of consent on you, not the platform. And a technical safeguard (like voice verification on professional clones) is not a consent process for your project. For instant clones, nothing on the page I read described a check that the voice belongs to the uploader, so the proof that you had permission has to come from you.
What should a voice cloning consent form cover?
A text message saying "sure, go ahead" is weak evidence and says nothing about limits. A one-page written agreement is better. Here is an untested template; have a lawyer adapt it for your country and your use.
VOICE CLONING CONSENT
Voice owner: [NAME, contact]
Requester: [NAME / COMPANY, contact]
Date: [DATE]
1. I agree that recordings of my voice may be used to create a synthetic voice model using [TOOL / PROVIDER].
2. Permitted uses (list every one): [E.g. narration of training videos for COMPANY]
3. Not permitted: use in advertising for third parties, political content, content that suggests I said things I did not approve, use outside the list above.
4. Disclosure: audio made with the model will be labelled as AI-generated [WHERE / HOW].
5. Approval: I will [review / not review] scripts before release. Changes to the script list require my written approval.
6. Term: this permission lasts until [DATE] or until I withdraw it in writing, whichever is first.
7. Withdrawal and deletion: if I withdraw, the requester will stop new use within [N] days and delete the voice model and source recordings, and confirm in writing.
8. Payment: [AMOUNT / NONE]. Compensation is [one-off / per use].
9. Where recordings and models are stored, and who can access them: [DETAIL]. The provider's own terms may give it rights over uploaded recordings and models that I cannot cancel; I have read them: [YES / NO].
10. Governing law and who to contact with questions: [DETAIL].
Signature (voice owner): ______ Signature (requester): ______
Points that people forget: a term and a way out (voices outlive jobs and relationships), a disclosure line so the owner knows the audio will be labelled, and a list of forbidden uses so "narration" doesn't quietly become "ads". If the voice owner is a minor, a deceased person, or an employee being asked by their employer, stop and get proper advice: the consent is not simple in any of those cases.
What belongs on a pre-clone checklist?
Run this before you create the voice.
| Check | Done |
|---|---|
| The voice is mine, or the owner signed a written consent covering this exact use | |
| The use is listed in the consent; no extra uses planned | |
| The person is not a political candidate or elected official | |
| The audio will not claim or imply the person said something they didn't approve | |
| Listeners will be told it is AI-generated, in the audio or the first line | |
| Source recordings were made with the owner's knowledge | |
| I know where source files and the voice model live, and who can access them | |
| A deletion date and process is agreed | |
| I read the provider's current use policy and terms on the day, and saved the date | |
| Local law on voice, likeness, privacy and consumer protection was checked with a professional |
If you can't tick the first, second and fifth rows, don't clone.
When is using your own voice still risky?
Your own voice is the clean case, but there are still traps. A cloned voice of yours can be used to fool a voice-based verification at a bank or in a company, so don't rely on your voice as a secret. Think about whether you want an always-available clone of yourself existing at all, who gets access to your account, and what happens if the account is compromised. Use two-factor authentication on the account and keep the number of people with access small.
Also separate the recording from the model: the first contains everything you said, including names and background sounds, and the second can say anything. Delete the source recording once the clone is made if you don't need it, but note that the terms grant ElevenLabs a perpetual, irrevocable licence to your content and voice models, so read what that means for you before uploading.
How do you use a cloned voice on a phone agent responsibly?
If the clone speaks to callers, the disclosure rule bites hardest. A cloned voice sounds like a person, so be more explicit than usual. Put "automated assistant" in the first sentence, don't let the agent claim to be the human whose voice it uses, and give callers a path to a real person. The calling and recording rules in India are covered, with their limits, in call recording consent and AI disclosure in India. The prompt patterns for the call itself are in voice agent call scripts for appointment booking and voice AI prompting for VAPI and ElevenLabs.
If you want to try ElevenLabs, my referral link is here. Some links on this page are affiliate links; they cost you nothing and I only link tools I'd mention anyway. It isn't the only option: other providers offer voice cloning, with their own policies, and a stock or designed voice avoids every consent question in this post. For many projects that's the better choice.
Reasons to say no
Decline the job, or the clone, when:
- The voice owner is unavailable to sign anything and you'd be relying on "they won't mind".
- The purpose is to make someone appear to endorse, say or confirm something they haven't.
- The listener is meant to believe they're speaking to a human.
- The voice belongs to a public figure and you have no direct permission.
- You can't say where the recordings are or when they'll be deleted.
A synthetic voice you designed from scratch, or a licensed stock voice, carries none of these problems.
The short version
Get consent in writing and scope it. Label the audio as AI. Keep the model and the recordings under control, with an end date. Reread the provider's policy on the day you start, because these pages change: the use policy I read was updated in August 2026 and the terms in March 2026. And for anything beyond a policy question, such as whether a particular use is lawful where you live, ask a professional rather than a blog.
