Voice cloning laws are being tested in court, and one of the most recent cases sits squarely in dubbing. In April 2026 twenty-five French dubbing actors obtained the removal of 47 AI-generated models that reproduced their voices without consent and without payment, from a platform based in the United States. Among the voices recognised were the French voices of Harrison Ford, Julia Roberts and Angelina Jolie. Their lawyer put it plainly: a company based abroad is not beyond the reach of European law when it offers its services to the EU public.
Voice cloning has stopped being a purely technical matter and turned into a question of rights and contracts. From 2 August 2026, anyone distributing AI-created or AI-manipulated audio to the EU public must disclose it in a recognisable way. And using an actor’s voice to train a model requires their explicit consent and a defined usage licence. A release signed years ago for a dubbing session does not cover this scenario.
Why an Actor’s Voice Is Protected Under Voice Cloning Laws
A person’s voice is personal data, and when it can identify them it takes on the character of biometric data protected under the GDPR. In several European jurisdictions it is protected in the same way as a person’s image, so reproducing it without authorisation can breach personality rights. On top of that sit the neighbouring rights in the recorded performance, which belong to the performer.
The point most buyers miss is a different one. A dubbing contract authorises recording and distributing a performance, not training a voice cloning system that generates new lines in that same voice. These are two separate uses, with two separate authorisations. The real risk today is not technological but contractual, and it almost always comes from a licence written for a world where voice models did not exist.
What Changes Under the EU AI Act From 2 August 2026
The obligation comes from Article 50 of the EU AI Act, and it is best read across two deadlines.
The Two Dates to Note
From 2 August 2026, anyone using AI to create a deepfake, meaning audio or video content that convincingly reproduces a real person, must disclose it clearly and visibly. The duty applies even without any intent to deceive. For generative systems already on the market before that date, the technical duty to mark audio in a machine-readable format has been pushed to 2 December 2026 by the revision known as the Digital Omnibus.
The penalties are not symbolic, they reach up to 15 million euros or 3% of total worldwide annual turnover. And the rule follows the market principle: anyone producing content aimed at the EU public falls within scope, even when established outside Europe. For a distributor localising into several languages, that means responsibility travels with the content, not with the border.
Anyone weighing a localisation with synthetic elements does well to talk early with a studio that handles both the creative and the regulatory side, rather than discovering the constraints once the work is done. It connects to what we covered on synthetic dubbing and when a human voice is genuinely needed.
What Voice Cloning Laws Require From Whoever Commissions a Localisation
Before authorising the use of a cloned voice, a few checks reduce the legal risk:
- Explicit written consent from the actor, tied to that specific use
- A defined scope of use, covering purpose, duration, territory and channels of distribution
- Rights to modify and reuse, to establish whether the model can generate new material
- Revocability of consent and termination conditions
- Traceability of the files used to train the model
Whoever commissions the work remains jointly responsible with the studio and cannot fall back on not knowing if something is out of line.
At RED Audio the voice stays a job done by people, with a network of more than 500 performers and release agreements that set out clearly how, where and for how long a voice can be used. Transparency about consent and use is not red tape, it is what keeps a project clear of disputes as voice cloning laws take effect. Anyone planning a localisation for the international market can define these terms from the outset with our team, through the Contact page.
Frequently Asked Questions
Do you need the actor’s consent to clone their voice with AI?
Yes. A voice is protected as personal data and as a likeness, and the neighbouring rights in a performance belong to the performer. Cloning an identifiable voice without explicit written consent invites takedown requests and damages claims, as the case of the French dubbing actors in April 2026 showed.
Does an old dubbing release cover training a voice model?
No. Authorising the recording and distribution of a performance is not the same as authorising that voice to generate new lines with AI. You need specific consent and a licence for training and for using the model.
What changes in practice from 2 August 2026?
Anyone distributing an audio or video deepfake to the EU public must disclose it visibly, even with no intent to deceive. The machine-readable technical marking, for systems already on the market, applies from 2 December 2026.
Who is liable if synthetic content is not disclosed?
Responsibility is shared along the chain. The party commissioning the localisation answers alongside the studio and cannot excuse itself by not knowing the obligations.
Do the rules apply to companies outside the European Union?
Yes. The AI Act follows the market principle: if the content is aimed at the EU public, the obligations apply even to companies established abroad. Penalties reach up to 15 million euros or 3% of worldwide turnover.
