Douglas Delannoy
Focus No. 230 October 2026, 11 min read, Topic Technology

Are actors expendable in the age of AI?

Face, voice, performance: AI is shaking up their rights.

Douglas Delannoy
Behind every performance, an artist.

The facts · The Lara Croft case

Lara Croft speaks with a cloned voice

In August 2025, an update to Tomb Raider IV-VI Remastered gave Lara Croft new lines spoken by an AI-cloned version of the voice of Françoise Cadol, her French voice since 1996, without her consent.

  1. CloneNew lines spoken by an AI-generated copy of her voice.
  2. ChallengeAn official court officer's report, then formal notice from her lawyer. The same alarm was raised by her Brazilian voice, Lene Bastos.
  3. WithdrawOn 19/09/2025, the publisher Aspyr removed all AI voice content.
According to her lawyerThe voice is protected as an attribute of personality, as biometric data (GDPR) and as a performance. His client had exercised her right to object.
Sources

ActuIA, 15/09/2025 – PC Gamer, 24/09/2025 – GamesRadar, 09/2025 – Time Extension, 09/2025

The facts · The phenomenon

3 seconds of recording to clone a voice

Their face, their voice and their performance become data, endlessly reusable.

3 sThe VALL-E model imitates a voice from a 3-second clipMicrosoft · 01/2023
100+AI-generated voices offered to audiobook publishers, including French voicesAudible · 05/2025
12Films and series dubbed with the help of AI, in a first trialPrime Video · 03/2025
From dubbing to audiobooksEvery recording an actor makes can become the raw material for a cloned voice.
Sources

Microsoft Research, VALL-E, 01/2023 – Audible, press release, 05/2025 – Gameblog, 07/03/2025

The facts · The profession sounds the alarm

4,000 actors sign ahead of the Césars

On 23/02/2026, three days before the 51st ceremony, an open letter backed by Adami called for an urgent legal framework.

Open letter of the 4,000, 23/02/2026“Cloning actors' voices without permission is becoming rife.” (translation)
€250offered to an actor for the AI use of his image, instead of 2 days' filming

#TouchePasMaVF

Since 01/2024, the SFA-CGT union and the association Les Voix have been mobilising dubbing actors behind this slogan, “Hands off my French version”.

Joint union letter

On 22/04/2026, SIA-UNSA, SFA-CGT, SNLA-FO and SNAPAC-CFDT called for the presumption of use to be adopted.
Sources

Adami, open letter of 23/02/2026 – joint union open letter, 22/04/2026 – 3DVF, 15/01/2024

Exploitation · On screen

A face digitised without consent: what does the law say?

A scan made during a shoot makes it possible to recreate the performer in other scenes, or even other films.

Image rights

Everyone, in a private or public place, has an exclusive right over their image as soon as they can be identified. It cannot be used without their express consent, ideally in writing (French Civil Code art. 9). This right does not expressly cover AI training, nor the performance itself.

Performers' rights

They protect the performance, but signing with an audiovisual producer counts as authorisation to fix, reproduce and broadcast it (CPI art. L212-3 and L212-4).

What is missing

Specific consent for a digital double. French law provides none for creating and then reusing a copy of a performer, unlike California or the SAG-AFTRA agreement.

Deepfake shared without consent: maximum penalties

France, general case1 year and a €15,000 fine
France, shared via an online service2 years and €45,000
Italy1 to 5 years in prison
Denmark (bill)No criminal penalty, removal and compensation
Everywhere, the victim can also take civil action, for removal of the content and damages.
Sources

French Civil Code art. 9 – CPI art. L212-3 and L212-4 – California, AB 2602 – French Criminal Code art. 226-8 (SREN law, 2024) – Italian law 132/2025, art. 612-quater of the Criminal Code – Folketinget, L 49

Exploitation · The performance

An avatar built from the acting of several performers

The law protects a performance, not a character that AI might generate from the cloned acting of several actors.

What the law says

Fixing, reproducing or broadcasting a performance requires the performer's written authorisation (CPI art. L212-3). Their moral right protects the integrity of their performance, with no time limit (CPI art. L212-2).

The Tilly Norwood case

An “actress” generated entirely by AI, unveiled in Zurich in 09/2025. SAG-AFTRA condemned a character “trained on the work of countless professional performers”, without permission or pay.

How can these rights be defended?

Without access to the training data, it is impossible to prove which performances fed the avatar.
Sources

CPI art. L212-2 and L212-3 – SAG-AFTRA, statement of 30/09/2025 – Variety, 30/09/2025

Exploitation · Dubbing

1 cloned voice for every language

Platforms are testing AI-created dubbing on titles that were never dubbed. You can guess what comes next…

Prime Video

03/2025, a trial on 12 films and series, in English and Latin American Spanish. Professionals check the quality, according to Amazon.

Audible

05/2025, AI translation of audiobooks announced, from English into French, Spanish, Italian and German.

#TouchePasMaVF calls for

  • Transparency for the public.
  • The right to authorise or refuse training.
  • Public funding tied to employing artists.
  • Quotas for AI-free works.
The questionWhat becomes of dubbing work?
Sources

Gameblog, 07/03/2025 – Audible, press release, 05/2025 – 3DVF, 15/01/2024

Exploitation · The weight of dubbing

12,500 jobs in France, a market of ≈ $4bn worldwide

In 2023, dubbing involved 110 companies in France, mostly in the Paris region, for more than €210m in gross salaries.

Who works in dubbing in France?

Freelance actors and technicians (intermittents)7,397
Permanent staff3,116
Adapters and writers1,811

Dubbing revenue

World, dubbing and voice-over (2024)≈ $4.2bn
France, dubbing (2023)≈ $0.7bn (€650m to €700m)
NoteThe figures are estimates, the industry's own for France and a research firm's for the world, covering different scopes.
Sources

Audiens DATALAB for LES VOIX and SFA-CGT, 16/05/2024 – Sacem, 05/2024 – Market.us, 20/01/2025 – average ECB rate 2023 (€1 = $1.081)

Exploitation · Audiobooks, video games

320 days on strike to regulate digital doubles

A strike almost 3 times longer than the 2023 film and TV strike, ended by an agreement ratified in 07/2025 between SAG-AFTRA and video game publishers. It regulates digital replicas created by AI.

Length of SAG-AFTRA strikes

Film and television, 2023118 days
Video games, 2024-2025320 days

Consent

Informed consent for each use of a digital replica.

Suspend

Consent can be suspended during a strike.

Pay

Minimum rates, up to 7.5 times the base rate for a voice generated in real time.
And audiobooks?Audible offers publishers more than 100 AI-generated voices. When these rely on training and replace a performer, who consented, and who gets paid?
Sources

SAG-AFTRA, Interactive Media Agreement, 07/2025 – Bloomberg Law, 08/11/2023 – Audible, press release, 05/2025

The law · In France

A presumption of use passed unanimously by the Senate

Today, it is up to the artist to prove that their performance was used to train an AI. The presumption aims to reverse the burden of proof. But can the AI provider (OpenAI, Google, Mistral AI, ElevenLabs…) be taken at its word?

  1. 08/04/2026The Senate passes the bill unanimously.
  2. 22/04/20264 artists' unions call for it to be adopted.
  3. 02/06/2026Passed in committee at the National Assembly, by 29 votes to 8.
  4. 11/06/2026Blocked in the chamber by around 110 amendments.
  5. 01/07/2026A parliamentary report proposes instead a mandatory financial contribution from AI providers, paid to rightsholders.
LimitThe presumption targets training. But no tool today lets an artist check, without relying on the AI provider's own statements, that their voice was not used for training and then exploitation. Nor does it deal with imitating a voice or a face without copying a recording.
Sources

Senate, 08/04/2026 – joint union open letter, 22/04/2026 – AFP, 02/06/2026 – Livres Hebdo, 01/07/2026

The law · Close-up, the opt-out

The opt-out protects the recording, not the voice

On 11/03/2025, Adami and Spedidam declared that they opposed text and data mining for AI training (CPI art. L.122-5-3 and L.211-3 8°). A rule has been set, but the law requires a machine-readable opt-out for online content, and with no means of monitoring, compliance cannot be verified to date.

Adami

For the rights entrusted to it by its members, subject to rights already transferred to third parties. On paper, mining their sound, audiovisual and film performances will require prior authorisation.

Spedidam

For the recordings of artists who have brought it their rights through its statutes or by express mandate, whatever the type of AI. It requires prior authorisation and appropriate remuneration, with no tool to date to check that they are respected.

What they do not cover

Rights already transferred, often to the producer (CPI art. L212-4), who must then exercise their own right to opt out (L211-3 8°). Mining for scientific research, which cannot be opposed (L122-5-3, II). Models already trained.
Sources

CPI art. L.122-5-3, L.211-3 8° and L212-4 – Directive (EU) 2019/790 art. 4 – Adami and Spedidam, press releases of 11/03/2025

Oversight · Who checks?

A right to say no, with no way to check

Today

Artists and CMOs have no tool to find out whether their performances were used to train an AI. The AI Act only requires a public summary of training data, with no detail per performer (art. 53).

What is needed

An independent third party, equipped to check the data used. The presumption of use, to reverse the burden of proof.

Blind spot, the imitated voice

A voice “in the style of” escapes the opt-out. In the United States, the court ruled out copyright in the voice itself (Lovo).

Blind spot, the digitised face and body

A deepfake is an image, video or voice faked by AI that makes a person say or do something they never said or did. Motion capture records a performer's movements and expressions, and these digital copies can animate other characters. Their reuse falls under the contract, image rights, performers' rights and the Criminal Code (art. 226-8), not under the opt-out.
Sources

AI Act art. 53 – Senate, presumption of use bill, 08/04/2026 – SDNY, Lehrman v. Lovo, 10/07/2025 – French Criminal Code art. 226-8

The stakes · The grey areas

3 grey areas

The voice “inspired by”

Without copying a recording, there is neither a neighbouring right nor an opt-out. In the United States, the court ruled out copyright in the voice (Lovo, 07/2025).

The contract signed before AI

Does an old transfer of rights cover uses that did not exist? Adami excludes rights already transferred from its opt-out.

After death

A voice or face recreated after death. Who consents, and for how long?

How long does protection last?

Performer's economic rights50 years after the performance
Published sound recording70 years after publication
Danish bill, voice and appearance50 years after death
Performer's moral rightNo time limit
Sources

SDNY, Lehrman v. Lovo, 10/07/2025 – Adami, 11/03/2025 – CPI art. L211-4 and L212-2 – Folketinget, bill L 49, 08/10/2026

Outlook · Levers for change

Consent, from capture to remuneration

Who can act? Law: the legislator, in France or the EU. Contract: agreements between producers, platforms, unions and CMOs.

  1. Consent

    A written agreement, specific to each AI use and revocable, modelled on the SAG-AFTRA video game agreement.
    Contract + law
  2. Prove

    Adopt the presumption of use and open training data to independent audit.
    Law (France)
  3. Protect

    Create a right over one's voice and appearance, like the Danish bill, beyond the recording alone.
    Law (France, EU)
  4. Pay

    Remuneration for each reuse of a digital replica, paid through CMOs.
    Collective management
Sources

French and European framework, rules differ outside the EU – SAG-AFTRA, 07/2025 – Senate, 08/04/2026 – Folketinget, L 49

Outlook · The law forecast · outlook at 30/10/2026

Protections moving forward here and elsewhere

FranceBlockedPresumption of use blocked in the National Assembly
European UnionIn forceAI Act: deepfakes labelled since 08/2026
DenmarkUnder discussionRight over one's voice and appearance, bill tabled on 08/10/2026
ItalyIn forceDeepfake shared without consent, voice included: 1 to 5 years in prison
United StatesUnder discussionNO FAKES Act (voice, likeness) passed in committee
CaliforniaIn forceDigital replicas regulated in contracts since 2025
ChinaIn forceLabelling of AI content, audio included
Sources

National Assembly – AI Act art. 50 – Folketinget, L 49 – Italian law 132/2025 (art. 612-quater of the Criminal Code) – S.4591 (United States) – California, AB 2602 – CAC (China)

Test yourself on this topic4 questions in Quiz No. 1 are about this publication: voice cloning in seconds, opting out and voice imitation, data mining for research, the presumption of use.
Take the quiz →

Read next All publications →