AI voice cloning consent: what ElevenLabs, OpenAI and Microsoft make you prove

AI voice cloning consent is not one rule but four different checks, according to the vendors' own documentation as we read it on October 25, 2026. ElevenLabs Professional clones accept your own voice only, even with someone else's consent. OpenAI and Microsoft take another person's voice, but only from that speaker's recorded statement, and both gate the feature behind approval.

That difference decides whether your project is possible at all. If you want to clone a client's, a colleague's or a narrator's voice, in our view the tool you pick matters before price or sound quality does. Below: what each vendor asks for, the exact steps, and the gaps where a vendor's page says nothing.

AI voice cloning consent takes four forms: a tick box, a recorded phrase, a verified voice, or an application

Upfynd lists 145 AI audio and voice tools as of October 2026, and the ones that clone voices do not agree on what "consent" means. This table covers the four vendor routes whose documentation we read.

OpenAI also offers a prompt-based route that needs no recording at all, covered below. It cannot copy a real person.

Can you clone someone's voice with their permission?

It depends on the tool, and the answer for ElevenLabs is two different ones. For its Professional Voice Clone, the ElevenLabs guide says: "For now, we only allow you to clone your own voice." Its FAQ goes further: "Even with their consent, you cannot clone someone else's voice." The same FAQ says: "All Professional Voice Clones require a verification process to confirm that the voice belongs to you."

The Instant Voice Clone is looser. The Instant clone page says that before saving you must "confirm that you have the right and consent to clone the voice." The page we read describes no separate identity check for that route. The ElevenLabs Prohibited Use Policy, updated October 9, 2026 by its own date, bans cloning a living person's voice "without their explicit and informed consent or other valid authorization."

In our view, that makes the Instant route a promise rather than a proof: the check is your own confirmation, and the policy is what holds you to it. The Instant clone page also says all generated audio "can be instantly traced back to the user responsible for the generation."

OpenAI and Microsoft take the opposite design. The speaker has to be on the recording, so a third party's voice can only be cloned if that third party records the consent themselves.

How to clone your own voice with ElevenLabs

Both routes work from your own recordings. These are the steps from the ElevenLabs docs, read October 25, 2026.

Instant Voice Clone, the quick route

1. In the dashboard, open Voices and click the plus icon.
2. Choose Instant Voice Clone.
3. Upload or record audio. The page recommends about 1-2 minutes of clear speech and advises against going past 3 minutes.
4. Name and label the clone, then confirm the consent statement and save.
5. Open My Voices and click Use voice.

Each file must be at least 5 seconds and the files together at least 10 seconds. The page recommends MP3 at 192 kbps or higher and says uncompressed formats like WAV usually do not improve quality. One speaker, no reverb, no long silences. The docs also say you cannot download a clone as a standalone file; it stays in your ElevenLabs account.

Professional Voice Clone, the slow route

The Professional guide asks for at least 30 minutes of audio and recommends 2-3 hours, in one speaker and one delivery style. It does not support singing. After upload you read verification lines through your browser microphone, once each, in a tone similar to your samples. If verification fails you can retry after 24 hours or contact support.

The docs disagree with themselves on training time: the FAQ estimates 3-6 hours and another section says 6-24 hours. We would plan for the longer one. Professional cloning needs the Creator plan or above, according to the ElevenLabs voice cloning overview; free and Starter accounts get no Professional slot there.

The OpenAI custom voice consent phrase, step by step

OpenAI's custom voices guide offers two routes. Prompt-to-voice is generally available to all API customers. Recording-based voices are "limited to eligible customers," who contact sales and then get a Voices tab under Audio.

For a recording-based voice, the speaker reads an approved phrase in a supported language. The English phrase is: "I am the owner of this voice and I consent to OpenAI using this voice to create a synthetic voice model." The guide adds that "Any divergence from the script will lead to a failure."

1. Fetch the supported phrases with a GET to `/v1/audio/consent_phrases`.
2. Upload the consent recording with a POST to `/v1/audio/voice_consents`, sending a name, a language and the recording. You get a consent ID back.
3. Create the voice with a POST to `/v1/audio/voices`, sending a name, an audio sample and the consent ID. Save the voice ID.

The consent recording and the sample must be the same speaker, and one consent can be reused for more voices by that speaker. Samples can be mpeg, wav, ogg, aac, flac, webm or mp4, up to 10 MiB. The guide gives two length figures: 60 seconds or less in one place, and 10-60 seconds in another, with at least 5 seconds of real speech. Aim inside both. Browser recordings labeled `audio/webm;codecs=opus` are rejected; use plain `audio/webm`.

Limits to know before you build on it: 20 recording-based voices per organization, versus 100 prompt-generated ones. Creating voices through this endpoint "isn't compatible with Zero Data Retention." Recording-based voices work with GPT-Live, the Text to Speech API, the Realtime API and Chat Completions with audio output. Prompt-generated voices work only with GPT-Live.

The prompt route has its own rules. The guide says not to imitate named people, public figures or fictional characters, and that children's voices are not allowed. Its responsible-use section also says to tell listeners the speech is AI-generated and to keep any provenance metadata or watermarks. The guide does not state a price for recording-based voices, so ask sales. For the API itself, Upfynd lists GPT-4o and OpenAI's API under LLMs and AI infrastructure.

Microsoft's consent statement names your company

Microsoft's personal voice overview, dated September 9, 2026, says "it's required that every voice be created with explicit consent from the user." You need a verbal statement plus a 5-90 second audio prompt. API access "is restricted to eligible customers and approved use cases," requested through an intake form. Professional voice fine-tuning is also approval-only, and its table says no unapproved use case is allowed.

Microsoft's consent page gives the English statement: "I [state your first and last name] am aware that recordings of my voice will be used by [state the name of the company] to create and use a synthetic version of my voice." The statement does double duty. The page says it is also used to verify that the voice talent is the same person as the speaker in the training data.

Two details trip people up. The voice talent name must match the person who recorded the statement, and the company name must match what was spoken. In the REST flow, the talent name, company name and locale cannot be changed later. If you run on Microsoft Azure, set those correctly the first time.

One listed cloning tool answers consent in a FAQ, not in a check

Murf AI is listed on Upfynd, and we state only what its voice cloning page contained when we read it on October 25, 2026, not what the product does behind the scenes. Its FAQ says "voice cloning itself is not illegal, but using voice duplication without consent can be considered illegal." Asked whether you can clone a celebrity voice, the answer begins "Yes," then warns that doing so without permission can lead to "intellectual property infringement or even identity theft."

The page content we read described no consent check or identity verification. Its marketing copy speaks of replicating "your own voice", but we found nothing on the page that enforces that. That differs from the ElevenLabs Professional route, which refuses a third party's voice outright. A page being silent is not proof that no check exists. In our view it is a reason to ask the vendor, in writing, what it verifies before you upload anyone else's audio.

A vendor's consent check is not legal clearance

These checks protect the vendor and prove a speaker agreed to a use. They do not tell you what the law lets you do with the result. Rules on voice and likeness differ by country and, in the US, by state, and we did not review any statute for this post. ElevenLabs' policy separately bars impersonating political candidates or elected officials without authorization, bars submitting voice recordings of anyone under 18 to create a voice model, and bars using biometric or voice data to identify or profile an individual. If your project touches advertising, a public figure or a minor, read those terms in full.

Synthetic audio is also heading toward disclosure rules. Our earlier post on EU AI Act watermarking requirements covers what its title says you must ship by December 2, 2026.

This is not legal advice. Vendor terms and documentation change, and we read these on October 25, 2026, so read the current terms before you build on any of this and take advice where real money rests on it. Quoted rules above are the vendors' own words; none is a statement of law.

Which route to pick

• Cloning your own voice for narration: ElevenLabs Instant for a quick test, Professional if you can supply 30+ minutes of clean audio and want the stricter ownership check.
• Cloning a client's or colleague's voice: OpenAI or Microsoft, where the speaker records the consent themselves, and plan for an access request. In our view the recorded statement is the cleaner paper trail.
• A new synthetic voice that belongs to nobody: OpenAI's prompt-to-voice, free to create, but GPT-Live only.
• A tool whose page describes no consent check: ask the vendor in writing what it verifies before you upload anyone else's audio.

To compare other tools, browse the AI Audio & Voice category. If you make a voice product and document your consent flow clearly, you can list it on Upfynd for free.