Cloning a Voice or a Face: What the Law Says

READ7 min

WORDS1,396

SECTIONS6

TYPEGUIDE

CHECKED31 AUG 26

Copyright asks who owns the output. That is the wrong question for a cloned voice or a synthetic face. The right one is older, and in some places it now has teeth.

TL;DR — THE SHORT VERSION

This is not copyright. It is the right of publicity — control over your own name, voice and likeness. Different law, different question, different answer.
Tennessee's ELVIS Act made voice a protected property right in 2024, and it covers a simulation of a voice, not only a recording of one.
There is no federal US law yet. The NO FAKES Act was reintroduced in May 2026 and is still a bill — treat any claim that it is law as wrong.
So protection is a patchwork. It depends on the state, and increasingly on where the viewer is rather than where you are.
Consent is the whole thing. Written, specific, and scoped — "for this project" is not the same as "forever, anywhere".
The dead are not automatically fair game. Several states extend these rights after death, for decades.

IN PLAIN ENGLISH

Copyright protects a thing you made. Right of publicity protects you — your face, your name, the sound of your voice.

That is why "I generated it, so it is mine" falls apart here. You can own a recording completely and still have no right to release it, because the person it sounds like has a say that has nothing to do with who made the file.

Two different questions, routinely confused

Who owns this audio file? Copyright. Answered by who owns what you generate — and usually the answer is "less than you think".COPYRIGHT

May I release something that sounds like a specific person? Right of publicity. Nothing to do with who made the file.PUBLICITY

You can lose on the second while winning the first. Owning the file is not permission to publish it.

The law that named voice specifically

What Tennessee did

Tennessee's Ensuring Likeness Voice and Image Security Act — the ELVIS Act — was signed on 21 March 2024 and took effect on 1 July 2024. It updated the state's existing publicity law for generative AI.

The part that matters: it treats voice as a protected property right, and defines it to include not just an actual recorded voice but a simulation of that voice. That closes the gap a cloning tool walks through — nothing of the original recording survives in the output, and until this it was unclear whether anything had been taken at all.

Enforcement runs both ways: a civil action, and criminal liability as a Class A misdemeanour.Tennessee ELVIS Act, signed 21 Mar 2024, effective 1 Jul 2024, amending the state's Personal Rights Protection Act. Widely reported as the first US law to name voice explicitly in the context of AI simulation. State law: it binds conduct connected to Tennessee, not the whole country. Checked 31 Aug 2026

Why one state matters more than it sounds

Nashville is there, so a large share of American commercial music has a Tennessee connection — and a law only has to reach your activity, not your address. Other states have since moved in the same direction, at different speeds and with different definitions.

The practical read: do not treat "my state has no such law" as a defence. Distribution is national, and the question is where the harm lands.

The federal law that does not exist yet

NO FAKES is a bill, not a statute

The NO FAKES Act would create a federal digital-replica right — a nationwide rule against producing or distributing an unauthorised digital replica of a person's voice or likeness. It was reintroduced on 20 May 2026 in both chambers with bipartisan sponsorship.

Reintroduced is the operative word. It has been introduced before and did not pass. As of this page's date it is not law, and a great deal of writing about it is phrased as though it were.NO FAKES Act reintroduced 20 May 2026 as S.4591 (Senate) and H.R.8915 (House); subsequently the subject of a Senate subcommittee hearing. Status as of 31 Aug 2026: pending, not enacted. Bills change and die; check the current status of both numbers before relying on this. Checked 31 Aug 2026

If it passes, the patchwork becomes a floor. Until then, your exposure depends on which state's law reaches you, and that is a question for a lawyer rather than a guide.

TAKEAWAY

Anyone who tells you AI voice cloning is either "totally illegal" or "totally unregulated" is wrong in both directions. It is state-by-state, moving, and the direction of travel is clearly toward more protection.

Almost every safe use of a real person's voice or face runs through consent, and most consent people collect is too vague to be worth having.

In writing. A verbal yes is not a record, and the person who gave it may not be the person you deal with later.□

Specific about use. "For this campaign" is a different grant from "for any future project".□

Specific about time. Permission that never expires is the clause people regret.□

Specific about training. Consent to use a recording is not consent to build a model from it. Say which you have.□

Revocable, with a stated process. What happens to the model and the output if they withdraw?□

THE CLAUSE PEOPLE FORGET

Training is a separate permission from use. A voice actor who agreed to record an advert did not thereby agree that their voice becomes a reusable synthetic asset — and this is the exact dispute that has driven most of the legislation on this page.

If your contract does not mention model training, assume you do not have it.

The dead, and the famous

Death does not reliably end these rights. Several US states extend right of publicity for decades after death, controlled by the estate. A historical figure is not automatically available, and "they cannot sue" is not the test — their estate can.

Fame cuts the other way from what people expect too. A well-known person has more commercial value in their likeness, not less, which is precisely what these laws protect.

The one genuinely safer path is a voice or face that belongs to nobody — a fully synthetic persona not modelled on a specific person. That is also the one most tools now offer by default, and it is the sensible default for commercial work.General description of US right-of-publicity practice, which is state law and varies in duration and in whether it survives death. Not legal advice and not a survey of all fifty states. Checked 31 Aug 2026

Before you publish a cloned voice or face

1 — Is this modelled on a real, identifiable person at all? If no, most of this page does not apply.□

2 — Do I have written consent, and does it name training as well as use?□

3 — Is the grant scoped in time and in purpose, rather than open-ended?□

4 — If the person is dead, have I checked whether their estate still holds the right?□

5 — Would a reasonable listener believe this is really them? That is the question the law tends to ask.□

6 — Am I disclosing that it is synthetic, where a listener would want to know?□

SOURCES AND SCOPE

Tennessee ELVIS Act — signed 21 March 2024, effective 1 July 2024 · NO FAKES Act — reintroduced 20 May 2026, S.4591 and H.R.8915, pending, not law. Checked 31 August 2026.

This is United States law and it is state-level and unsettled. Other jurisdictions handle image and voice through privacy, personality or data-protection law instead, and this page does not attempt to summarise them. It is not legal advice — it is a description of published positions with the dates attached, so you can check whether they still hold. Anything involving a real person's voice in commercial work is a question for a lawyer, and cheap to ask about compared with the alternative.

The through-line: you can own the file completely and still have no right to release it. The person it sounds like was never party to the copyright question, and that is the one that gets people.