Voice Cloning Consent: What Creators and Businesses Need to Know (2026)

Last updated: August 2026

Four separate, well-documented incidents since 2023 involve a voice being cloned in a way its owner did not expect, and in every one of them, the person or company doing the cloning could point to some form of permission. Stephen Fry had narrated audiobooks. Gayanne Potter had done paid work for a speech-tech company. Scarlett Johansson had been approached, and declined. ElevenLabs had a consent-statement step built into its product. Permission existed in some form in all four cases. What was missing was a properly scoped, documented, revocable agreement, and that gap is exactly what turned an ordinary recording session into a public dispute.

This guide is the practical companion to our legal guide to voice cloning, which covers what the law actually requires by state and country. This one covers something different: what consent needs to contain to actually hold up, a ready-to-use template for creators, a workflow for businesses managing this at scale, and the edge cases (minors, deceased voices, revoked consent) that most guides on this topic skip entirely.

What actually counts as valid consent?

“Sure, go ahead” is not consent in any legally useful sense. It does not say what the voice can be used for, for how long, whether the person can change their mind, or whether they are being paid for it. Guides and legal teams that have looked closely at synthetic voice agreements converge on the same five components, whether the context is a creator cloning their own voice or a company licensing a contractor’s, as detailed in a 2026 consent-checklist breakdown from AI infrastructure company WaveSpeed.

• Identity: whose voice is this, specifically, and does the person signing actually have the authority to grant rights over it? (This matters more than it sounds: a manager, agency, or platform account holder is not automatically the rights holder.)
• Scope: what is the clone allowed to be used for? “Internal training demo” and “public advertising” are not the same grant, and a permission that does not name the use case defaults to the narrowest reasonable reading, not the broadest.
• Duration and territory: how long does this permission last, and where does it apply? A clone approved for a single 2024 campaign does not automatically carry over into a 2026 product, and consent given in one country does not automatically travel with a global release.
• Compensation: is there a fee, a royalty, or a flat rate, and does it scale with how the voice is actually used? A rate that made sense for an internal e-learning module is not the same rate as one for a national ad campaign using the same clone.
• Revocation: what happens if the person changes their mind after the fact? Can they withdraw consent, and if they do, what happens to the model, the training data, and anything already published?

Miss any one of these and the agreement is incomplete, not just imperfect. The case studies below are what that looks like when it plays out in public.

What happens when consent is skipped or exceeded?

These four incidents span 2023 to 2025, and each one traces back to a different missing element from the list above, not a lack of any permission at all.

Stephen Fry, Harry Potter audiobooks (2023): Fry had narrated the audiobook series, recordings originally made for a specific, contracted purpose. In September 2023, at the CogX Festival in London, he played a clip of an AI-generated voice reading historical documentary narration, built entirely from his Harry Potter recordings, that he had never approved or been paid for, as reported by Variety. The failure here is scope: consent to narrate an audiobook does not extend to training a synthetic voice model for unrelated future use.

Gayanne Potter, ScotRail (2025): Potter recorded material for speech-tech company ReadSpeaker in 2021, understanding it would support accessibility and e-learning software. In May 2025, ScotRail introduced an AI train announcer, “Iona,” built from that same recording, for a use case she had never agreed to and was never asked about again, according to BBC News reporting. Potter told BBC Scotland: “It’s hard enough for people in the creative industry to sustain careers, but to be competing with a robotic version of yourself adds insult to injury.” Transport Scotland said it would introduce an alternative voice following the controversy. This is also a scope failure, four years after the fact, which is exactly why duration and use-case limits need to be written down at the time of the original recording, not assumed to expire on their own.

Scarlett Johansson, OpenAI’s “Sky” voice (2024): OpenAI approached Johansson in 2023 to voice its GPT-4o system. She declined. In May 2024, the company released a voice, “Sky,” that listeners and news outlets could not reliably distinguish from hers. Johansson said in a statement reported by CNBC: “When I heard the released demo, I was shocked, angered, and in disbelief that Mr. Altman would pursue a voice so eerily similar to mine that my closest friends and news outlets could not tell the difference.” OpenAI paused the voice after she retained legal counsel. This is the starkest case on this list: there was no consent at all, only a prior conversation and a refusal, which makes it a useful reminder that a declined request is not neutral ground.

Vacker v. ElevenLabs, Inc. (settled 2025): Voice actors Karissa Vacker and Mark Boyett, along with a publisher and an author, sued ElevenLabs (case 1:24-cv-00987), alleging their voices and copyrighted material were used without proper authorization. The parties reached a settlement in August 2025 and filed to dismiss the case in November 2025, making it the first resolved case among the wave of AI copyright lawsuits, according to court records on CourtListener and reporting tracked by AI Lawsuit Tracker. Settlement terms were not made public, but the case is a concrete signal that “we had a consent mechanism” is not, on its own, a complete legal defense if the scope and compensation terms behind it are contested.

Not every story here is a cautionary one. ElevenLabs launched an “Iconic Marketplace” in November 2025 specifically for licensing celebrity and public-figure voices with the rights holder’s direct participation and a revenue share, according to AlternativeTo’s coverage of the launch, which is roughly what a scope-and-compensation-complete consent process looks like when it is built into the product rather than bolted on after a dispute.

A ready-to-use consent template for creators

If you are cloning your own voice, or a collaborator’s, for a specific project, you do not need a law firm to draft something usable. You need the five elements above in writing, with both people’s names on it. Adapt the language below; the bracketed fields are what actually needs to change per project.

Keep the signed copy, not just an email thread. If you are using a tool like ElevenLabs that requires its own built-in consent statement before it will process a clone, that step confirms a recording was submitted with some form of authorization, but it is a floor, not a substitute for the document above. The platform’s check protects the platform. This document protects you.

How should businesses manage voice cloning consent at scale?

The creator template above breaks down once more than two people are involved, which is the normal case for a business: an employee whose voice powers an IVR system, a contracted voice actor whose recordings train a customer-facing assistant, or a vendor supplying a synthetic voice product to your company. Three additions matter at that scale.

• A named consent owner. One person or role (not “legal,” not “whoever set this up”) is responsible for knowing where every signed consent document lives and being able to produce it on request. Without this, “we have consent somewhere” tends to mean nobody can find it when it matters.
• Vendor flow-down language. If a vendor supplies you with a synthetic voice, their contract with you needs to explicitly state that they hold valid, scope-matched consent from the underlying voice source, and that this warranty survives the contract. Otherwise you have no recourse if their consent turns out to be incomplete, and you are the one facing the public dispute, not them.
• An approval workflow before every new use case. The ScotRail and Fry cases both happened because a recording collected for one purpose was later reused for a different one without going back to the original person. A one-line internal rule, “any new use case for an existing voice clone requires a fresh sign-off, not just a database lookup,” would have prevented both.

If your team is choosing a voice cloning platform partly on this basis, our roundup of voice cloning tools covers pricing and safe-use practices tool by tool, and is worth checking before committing to one for high-volume commercial use.

What edge cases do most consent guides skip?

Nearly every consent checklist available in 2026, including the ones this article drew on for the five-element framework above, stops at the standard adult-creator-to-business scenario. Three situations come up often enough in practice that they deserve their own answer.

Minors. A parent or legal guardian must provide consent on behalf of anyone under the age of legal majority in the relevant jurisdiction, and that consent should be reviewed and renewed as the minor approaches adulthood, not treated as permanent from a single signature obtained when they were a child. Several of the state laws covered in our legal guide extend right-of-publicity protections without a general “unless a parent already agreed once” exception, so treat this as the strictest case, not the easiest one.

Deceased individuals. Voice rights do not automatically end at death. Several US states recognize a post-mortem right of publicity, meaning an estate, not the general public, controls whether a deceased person’s voice can be cloned, and for how long that control lasts varies significantly by state. If a project involves recreating a voice belonging to someone who has died, the estate or its named rights holder is the party whose consent is required, and that consent should follow the same five-element structure as any living person’s, with the estate substituting as the identity-and-authority party.

Revocation after publication. This is the gap almost nobody addresses in writing: what actually happens when someone withdraws consent after content using their clone is already live? The honest answer has three parts, and all three should be in the original agreement, not decided in the moment: whether existing published material gets pulled or is allowed to stand, whether the underlying model and training data get deleted, and what timeframe is reasonable for stopping new use (the consent template above suggests 30 days as a starting point, adjust to what is actually workable for your production pipeline).

How does this connect to the law?

A properly scoped, signed consent document is good practice everywhere, but it is not the same thing as legal compliance, and the two can diverge by jurisdiction. Six US states now have specific voice-cloning statutes with real penalties, and the EU AI Act’s transparency rules take effect August 2, 2026, requiring disclosure when audio is AI-generated regardless of whether the underlying voice was properly licensed. Our state-by-state and country legal guide covers that jurisdictional detail in full; treat this article as what to put in the paperwork, and that one as whether the paperwork is enough where you operate.

Sources

• Variety, “Stephen Fry Says AI Stole His Voice From Harry Potter Audiobooks,” 2023, variety.com, retrieved 2026-08-03
• BBC News, “‘Stop using my voice’ – New train announcer is my AI clone,” 2025, bbc.co.uk, retrieved 2026-08-03
• CNBC, “Scarlett Johansson says OpenAI ripped off her voice after she told the company not to use it in software,” May 20, 2024, cnbc.com, retrieved 2026-08-03
• CourtListener, “Vacker v. ElevenLabs, Inc., 1:24-cv-00987,” docket, courtlistener.com, retrieved 2026-08-03
• AI Lawsuit Tracker, “Vacker v. ElevenLabs, Inc.,” case summary, ailawsuittracker.com, retrieved 2026-08-03
• WaveSpeed, “How Can You Get Consent for AI Voice Cloning? A Checklist,” 2026, wavespeed.ai, retrieved 2026-08-03
• AlternativeTo, “ElevenLabs launches a new ‘Iconic Marketplace’ for licensing AI generated celebrity voices,” November 2025, alternativeto.net, retrieved 2026-08-03

We verified each case above against the specific source cited. Settlement terms in the Vacker v. ElevenLabs matter were not made public; we report only what court filings and tracked case summaries confirm, not speculated terms.

Frequently Asked Questions

Is verbal consent enough to clone someone’s voice?

Not in any way that holds up if disputed. Verbal agreement does not document scope, duration, compensation, or revocation terms, all of which matter once a clone is used beyond what the person actually pictured when they agreed.

What are the five elements of valid voice cloning consent?

Identity (whose voice, and who has authority to grant rights over it), scope (what it can be used for), duration and territory (how long and where), compensation, and revocation (what happens if consent is withdrawn).

Can I clone my own voice without anyone else’s consent?

Yes. Consent requirements exist to protect other people’s rights over their own voice. Cloning your own voice with your own recordings does not require a third party’s permission, though platform terms of service still apply.

What happens if someone withdraws consent after their clone has already been used publicly?

This should be defined in the original agreement, not decided after the fact. At minimum, expect to stop all new use within a set timeframe and address whether already-published material is pulled or allowed to stand, and whether the underlying voice model is deleted.

Do I need special consent to clone a minor’s voice?

Yes. A parent or legal guardian must provide consent on the minor’s behalf, and it is worth revisiting that consent as the minor approaches adulthood rather than treating one early signature as permanent.

Can you legally clone the voice of someone who has died?

It depends on the jurisdiction. Several US states recognize a post-mortem right of publicity, meaning the person’s estate, not the general public, controls whether their voice can be cloned and for how long. Consent should be obtained from the estate’s designated rights holder.

Does a platform’s built-in consent check (like ElevenLabs’ consent statement) protect me legally?

It confirms a recording was submitted with some form of authorization, but it does not verify scope, compensation, or revocation terms between you and the voice’s owner. Treat it as a baseline the platform requires, not a substitute for your own signed agreement.

What’s the difference between this guide and your legal guide to voice cloning?

This article covers how to obtain and document consent properly, regardless of jurisdiction. Our legal guide covers what specific laws require by US state and country, including where voice cloning carries statutory penalties.

Practices and platform policies referenced here change frequently. Always confirm current terms directly with the relevant platform or a licensed attorney before relying on this guide for a specific project.

About the author

Richard Johnson

Richard Johnson is an AI specialist with over five years of experience guiding large organizations through AI adoption, across more than 100 customers. He founded CognitiveFuture to research and compare AI tools across design, development, writing, research, voice and business, cutting a crowded, fast-moving market down to the right choice for the job in front of you.