AI celebrity voice generators and the law: ELVIS Act, NO FAKES, and what to do instead

Tennessee made it actionable to distribute a tool whose primary purpose is producing an identifiable person's voice without their consent. Not to use one. To distribute one. That provision sits inside the ELVIS Act, signed in March 2024, and it is the clearest signal of where this is heading: the law is moving up the chain, from the person who posted the clip toward the person who built the machine.

On 22 June 2026 the Senate Judiciary Committee unanimously advanced the NO FAKES Act, which would create a federal right against unauthorised digital replicas of a person's voice or visual likeness. Unanimously, and bipartisan. That does not happen to bills nobody intends to pass.

If you are searching for a celebrity AI voice generator, this is the context you are searching inside. Here is what is actually settled, what is not, and what to do instead.

This summarises reported legal developments. It is not legal advice.

What the law already says

Right of publicity exists and predates AI. In most US states, using a person's identity for commercial purposes without permission is already actionable. Voice has been part of that in significant cases for decades. AI did not create this right; it made it cheap to infringe.

The ELVIS Act made voice explicit. Tennessee's Ensuring Likeness, Voice, and Image Security Act updated state law to say plainly that a person's voice is protected, including digitally generated imitations that sound like them. Its notable move is the tool provision: a private right of action against distributing technology whose primary purpose is producing a particular identifiable individual's voice or likeness without consent.

NO FAKES would federalise it. The bill creates a federal right to control digital replicas of voice and likeness. Two details matter. It is not limited to celebrities, so this is not only a famous-person issue. And as of its June 2026 committee advance it is a bill, not law, so nothing about it is settled yet.

Do not confuse it with TAKE IT DOWN. That federal act addresses non-consensual intimate imagery. Different scope, different mechanism, frequently muddled together in coverage.

Where that leaves the tools

A generator that advertises specific named voices is selling the thing the ELVIS Act names directly. Whether it is enforceable against a given tool in a given state is a question for courts, and courts are getting the question.

For you as a user, the exposure is simpler than the legislative picture:

Publishing is the trigger. Generating a private curiosity and publishing a track are different acts with different consequences. Almost every real problem starts at distribution.

Commercial use multiplies it. Right of publicity is fundamentally about commercial exploitation of identity. Monetising is the fastest route from novelty to claim.

Platforms enforce independently of the law. Content matching on streaming services and video platforms does not wait for a court. Takedowns, channel strikes and distributor blacklisting happen on the platform's own timetable, and monetisation requires holding 100 percent of the rights regardless of what any statute eventually says.

Disclaimers do not fix it. Labelling a track "AI cover, not the real artist" addresses confusion, which is only one part of the claim. It does nothing about unauthorised commercial use of identity.

Why the whole approach is a creative dead end anyway

Set the law aside for a moment, because there is a better argument.

Imitating a specific famous voice caps your ceiling at "sounds a bit like someone else." The best possible outcome is a good impression. That is a novelty format with a short shelf life, and it makes your work permanently derivative of a person who did not agree to be involved.

The interesting thing generative music can do is the opposite: produce a voice that has never existed and make it consistent enough to become recognisable as yours. That is a real asset. Nobody can send a letter about it, no platform matches it, and it appreciates as you use it.

What to do instead, concretely

Build a persona instead of a clone. Suno's personas fix a vocal identity across songs. Same practical benefit people want from a cloned voice, which is consistency and recognisability, with no real person attached. How voice and personas work is the practical guide.

Describe the sound, never the singer. This is the substitutive move and it works better than people expect. A named artist in a prompt is a shortcut for a bundle of attributes: vocal texture, era, production, arrangement. Name the attributes and you get the texture without the identity.

Instead of naming a performer, write what you actually mean:

breathy female lead vocal, close-mic intimacy, slight rasp on sustained notes,
1970s analog folk production, fingerpicked acoustic guitar, upright bass,
warm tape saturation, 92 BPM

That gets you a specific sound. It also gets you a sound that is yours, and it does not depend on any court ruling. Our entire style catalogue is built on this principle: 1,960 recipes written as instruments, era, production and vocal character, with no artist names anywhere. That is a deliberate design decision, not a limitation, and the reasons are the same ones a rights holder would give.

Use your own voice. If you want a specific human voice in your music and you have one, that is the one you unambiguously have rights to. Suno supports recording and using your own voice on paid plans.

Get permission if you truly need a specific person. For a real project with a real budget, licensing a voice is a normal transaction. Several artists have licensed their voices for AI use deliberately.

The trajectory to plan around

Three trends, all pointing the same way.

State laws are proliferating and voice-specific provisions are being added rather than removed. Federal legislation has cleared a committee unanimously. And the industry is converging on licensing as the settlement pattern, which structurally favours consent-based use.

Nothing in that picture suggests unauthorised voice cloning gets easier. Build the habit that survives it, which is describing sound rather than borrowing identity.

FAQ

Is using an AI celebrity voice generator illegal? Generating privately and publishing commercially are different acts. The second implicates right of publicity, and voice-specific laws like the ELVIS Act have made that explicit.

Is the NO FAKES Act law? Not as of its unanimous committee advance on 22 June 2026. It is a bill with momentum.

Does it only protect celebrities? No. The proposed federal right would apply to everyone, not only public figures.

Can I post an AI voice cover if I do not monetise it? Lower exposure, not zero. Platform enforcement and right-of-publicity claims are separate from whether you earned money.

Does saying "this is AI" protect me? It reduces confusion, which is one element among several. It does not authorise commercial use of someone's identity.

How do I get a recognisable singer legitimately? Build a persona, use your own voice, or licence a real one. All three are available and none require a court to bless them.