AI Voice Cloning and Dubbing of Creator Content: Rights and Disclosure Under the EU AI Act

What is AI voice cloning of creator content?

AI voice cloning is when an AI system is trained to reproduce a specific person's voice and is then used to make that voice say something the person never actually recorded. In influencer marketing, the most common form is AI dubbing: a brand takes a creator's existing ad read or video and uses AI to generate a cloned version of the same voice in another language, instead of recording a new local creator or hiring a human voice actor. The result sounds like the original creator speaking German, Spanish or French — but none of it was actually spoken by that person.

It is a real, increasingly common production decision: one recording can, in theory, be repurposed across several language markets without reshooting with local creators. But that decision triggers two separate questions that get conflated in practice — and both need to be resolved before the content goes live.

Two separate duties: consent and disclosure

The rights question: do you even have permission to clone the voice?

The first question is contractual: has the creator given you permission to use their voice for AI cloning? A general usage-rights grant over "the content" in a standard influencer contract does not automatically cover the right to train an AI model on the person's voice and generate new, synthetic speech with it — the same logic that applies when brands consider using creator content for AI training generally (see our article on licensing creator content for AI training). Voice cloning is a distinct use, and it deserves its own explicit clause.

The disclosure question: do you have to tell the viewer the audio is AI?

The second question is independent of the first: do you have to disclose to the public that the audio is AI-generated? The EU AI Act (Regulation (EU) 2024/1689) answers that clearly, yes, from 2 August 2026 — and it applies whether or not the creator consented to the voice cloning. Consent determines whether the use is lawful toward the creator. The disclosure duty determines whether the viewer is misled. One does not substitute for the other.

The EU AI Act, Article 50(4): what does it actually require?

Article 3(60) of the AI Act defines a deepfake as "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful." An AI-cloned dub of a real creator's voice fits this definition directly — sounding authentic is the entire point of the exercise.

Article 50(4) requires the deployer — the party using such an AI system — to disclose that the content has been artificially generated or manipulated. Per the European Commission's own guidelines and Danish legal analysis (Kromann Reumert, 2026), the disclosure must:

• be visible or audible at the moment of first exposure — not buried in a caption or a linked page
• for content that runs over time (video or audio), repeat "at the start and periodically throughout", not just once at the beginning
• be clearly identifiable — for image/text content, guidance references a minimum 4.5:1 contrast ratio; for audio, the requirement is an audible notice a listener can understand without using a detection tool

A hidden, machine-readable mark embedded in the audio file itself (see the next section) does not satisfy this duty on its own — that mark is aimed at systems, not at the ordinary listener.

Provider and deployer duties are two different things

A brand that commissions AI dubbing of a creator's voice is, in this context, almost always the deployer — you are using the AI system for your own commercial purpose, regardless of whether you built it. The disclosure duty sits with you, not just with the vendor supplying the dubbing tool.

Does the creator's consent remove the disclosure duty?

No. This is the most common misconception in this area. A listener who recognises a voice assumes the person is actually speaking — that assumption does not change because the voice was cloned with full, paid, lawful consent. The AI Act's deepfake definition is about what the content does to the audience, not how it was produced. The Commission's own guidelines are explicit that the lighter regime available for evidently artistic or satirical content "cannot be used to override fundamental rights" — and an ordinary ad dub is neither satire nor fiction, so the full disclosure duty applies.

What should a voice-cloning clause in the contract cover?

This is Make Influence's operational recommendation, not a legal checklist — but it follows the same pattern as the rest of our contract checklist: a voice-cloning clause should stand on its own, separate from the general usage-rights grant, and should cover at minimum:

• Explicit, specific permission for voice cloning — not inferred from a general "usage rights over the content" clause
• Scope: which languages, which markets and which platforms the clone may be used on (consent to a German Instagram dub does not automatically cover a Spanish TV commercial)
• A time limit, parallel to the ordinary content usage-rights period
• Separate compensation for the voice cloning, distinct from the fee for the original recording
• A way to withdraw consent going forward, with a clear process for doing so

The same logic applies here as to exclusivity clauses and confidentiality clauses in the contract: the more specific the wording, the less room for a dispute later. See our breakdowns of exclusivity clauses and NDA and confidentiality clauses for the same principle applied to other contract terms.

The Danish angle: a new copyright over your own voice and body is coming

Denmark is separately working on its own, broader protection: a bill (L 111, 2025-26 parliamentary session), built on a political agreement reached by eight parties in June 2025, would add a new § 73a to the Danish Copyright Act. It would give every natural person — not only public figures — a copyright-like protection against realistic digital imitations of their appearance, voice or movements being made available to the public without consent, with an exception for satire and parody.

Precision matters here: as of the research for this article (23 August 2026), this remains a bill under parliamentary treatment, not a law in force. The expected entry-into-force date has itself moved between official statements (originally 31 March 2026, later 1 July 2026), and the proposal met opposition in February 2026 from, among others, the European video games industry and the Danish think tank Justitia. Do not treat it as an enforceable right yet — but be aware it is coming, and that it will apply to voice cloning specifically, independent of whether the AI Act's disclosure duty has also been met. The two regimes will complement, not replace, each other: the AI Act is about informing the viewer; the Danish copyright right will be about the person's own right to refuse the use in the first place.

Decision framework: AI-cloned voice or a human voice actor?

Example (hypothetical)

A Danish brand has a creator's original, Danish ad read AI-dubbed into German and Spanish to reuse the same video across three markets. Here is what it looks like when both duties are handled: the contract carries a standalone voice-cloning clause naming exactly those two languages and the platforms the video will run on, with its own payment for the cloning. Each of the two dubbed versions opens with a short, audible notice ("This audio is AI-generated") within the first few seconds, plus a matching visible text label if the video is also viewed without sound. The original Danish version is unchanged and needs no AI label, because that voice is genuinely the creator's own. The figures in this example are invented and illustrative only.

FAQ

Does the creator's consent remove the need to disclose that the audio is AI-generated?

No. Consent makes the use lawful toward the creator; it does not change the fact that the viewer must still be told the audio is artificially generated or manipulated, under Article 50(4).

Does the disclosure duty apply if a human voice actor dubs the content into another language instead?

No. Article 50(4) applies to AI-generated or AI-manipulated content. An ordinary human translation or dub with no AI involved is not covered.

Who carries the disclosure duty — the brand, the agency, or the AI tool's provider?

Typically whichever party deploys the tool for its own commercial purpose — in practice, usually the brand or the agency commissioning the dub, not just the software vendor.

What is the penalty for not complying with Article 50?

Fines of up to €15 million or 3% of global annual turnover, whichever is higher — for smaller companies, the lower of the two thresholds applies.

Do the same rules apply to a fully synthetic AI voice that isn't meant to represent a specific real person?

That's a different question — see our article on AI and virtual influencers, which covers synthetic personas rather than the cloning of a real person's voice.

What if someone else clones a creator's voice without their consent, not a brand they're working with?

That's a different, more serious problem — a third party's unauthorized deepfake, not a brand's consensual dubbing decision. See unauthorized AI deepfakes of an influencer for the legal recourse available in that case.

Make Influence's perspective

In our experience, the most common mistake is not deliberately ignoring the rules — it's assuming that "we already have usage rights to the content" from the original contract also covers AI voice cloning. It generally does not. Our recommendation to brands considering AI dubbing of creator content: resolve the rights question with a standalone clause before the recording even happens — not afterwards, once the dub is already made and due to go live tomorrow.