Is voice cloning legal in the US? Consent, publicity rights and AI labels

There is no single federal law that says voice cloning is allowed or forbidden. Instead, the legality of a cloned voice depends on three questions that come up again and again. Whose voice is it, and did that person agree? What is the cloned voice being used for, and could it mislead the people who hear it? And how is it delivered, for example in an advertisement, a phone call or a training video?

If the voice is yours, the use is honest and the audience is told when audio is generated, most of the legal risk falls away. Most problems start when one of those three answers is missing: a voice used without permission, content designed to make listeners believe something false, or calls placed with an artificial voice without the consent the rules require.

Consent is the line that separates a useful tool from a liability. When the person whose voice is cloned agrees, in a way you can later prove, most of the legal questions become questions about the content rather than about the clone itself.

Good consent has a few features in common. It comes from the person themselves, not from their manager or agency on their behalf. It is specific about what the voice will be used for. It is recorded, so there is evidence of who agreed, when and to what. And it can be withdrawn, with a clear way to delete the clone when the person leaves the company or changes their mind.

Employers should pay particular attention here. An employee who records training narration has not automatically agreed to their voice being cloned and reused after they leave. Write down what the clone covers, for how long and what happens on departure, and have the employee agree to it themselves.

The right of publicity protects a person's identity, such as their name, likeness and in many places their voice, against unauthorized commercial use. In the US it is mostly a matter of state law. Some states protect it through statutes, others through court decisions, and the scope differs: what counts as commercial use, whether protection continues after death, and whether voice is named explicitly.

Several states have updated or extended their laws in recent years to speak directly to voice and to digital replicas created with AI, and more proposals are under discussion at both state and federal level. Because the rules move and differ by state, the safest working assumption is simple: do not use anyone's voice or likeness in commercial content without their documented permission, whether they are famous or not.

Public figures deserve a special mention. Using a cloned voice of a celebrity, politician or other well known person invites publicity claims, and in political contexts can also raise election law questions. That is why many responsible tools block public figures entirely.

The Federal Trade Commission enforces the federal prohibition on unfair or deceptive acts and practices in commerce. In general terms, if a cloned voice is used to make consumers believe something false, such as that a real person endorses a product they never endorsed, or that a call comes from a bank, agency or relative when it does not, that use can be treated as deceptive. The FTC has also adopted rules aimed specifically at impersonation of government agencies and businesses and has publicly discussed the risks of AI-enabled impersonation of individuals.

On top of consumer protection law, impersonation used to obtain money or information can fall under fraud and identity theft laws at both federal and state level. None of that depends on the technology used; a cloned voice is simply a new way to commit an old offense. Endorsement rules also apply: if your cloned voice appears in an ad, the endorsement still has to reflect your honest opinion, and material connections still have to be disclosed.

Phone calls are the area where rules on artificial voices are most concrete. Federal rules on telemarketing and automated calls have long restricted calls that use an artificial or prerecorded voice, generally requiring the called person's prior consent for many types of calls, with stricter requirements for marketing. The Federal Communications Commission has stated that AI-generated voices count as artificial voices under those rules.

In practice this means that dropping a cloned voice into outbound phone campaigns brings the same consent requirements as any prerecorded message, plus any state rules on top. Voice clones are a far better fit for content that people choose to watch or listen to, such as training, recorded follow-up videos and course lessons, than for unsolicited calls.

Telling the audience when a voice is generated is good practice and, in a growing number of contexts, an expectation set by platforms, advertisers and regulators. Some laws and platform policies already require labels on certain AI-generated content, particularly in political advertising, and several social platforms ask creators to label realistic synthetic media.

Disclosure also protects you. A clear label makes it hard for anyone to argue that viewers were misled, and it keeps trust with an audience that will eventually learn how the content was made. Two practical forms of disclosure work well together: a visible label on the video or in the description, and machine readable provenance data attached to the file, such as content credentials, that records how it was made.

• The voice belongs to you, or to a person who agreed to the clone themselves, and that agreement is recorded.
• The content is honest. It does not imply endorsements, relationships or facts that are not true.
• The audience is told that the audio or video is AI-generated, with a visible label where it fits.
• No public figures, politicians or celebrities are cloned or imitated.
• Phone campaigns with any artificial voice follow the consent rules for automated calls.
• There is a way to delete the clone and an owner responsible for it.
• For advertising, regulated industries or anything unusual, a lawyer has reviewed the plan.

IdClone was built around the idea that the only clone worth making is a verified clone of you. Before a clone exists, the person in the recording passes a live selfie check and reads a spoken consent statement with a one time phrase, matched to their training footage. Teammates get their own invite and record their own consent. No identity documents are uploaded or processed.

• Politicians, public figures and celebrities are blocked, and impersonation, fraud, deception and adult content are forbidden under acceptable use.
• Every export can carry a visible AI-generated label and content credentials metadata.
• Each clone has a consent record and an audit log; Business adds consent record export and Enterprise adds audit log export.
• Recordings are used only to build and run your clone, never to train shared models, and you can delete a clone any time, which removes the model files.

The details are on consent and safety. These controls make compliance easier, but they do not replace advice on your specific use, especially in advertising and phone outreach.