AI voice cloning: how to protect your voice in 2026
Can your voice really be reproduced without your knowledge and sold online? That's exactly what happened in France to twenty-five professional voice actors in early 2026. How can you protect yourself when just a few seconds of audio are enough to create a convincing clone?
The VoiceDub / Fish Audio case: what really happened
At the end of January 2026, eight French dubbing actors — including voices familiar to the general public, such as the performer who usually dubs Julia Roberts or Angelina Jolie in France — sent a joint cease-and-desist letter to two American voice‑cloning platforms, VoiceDub and Fish Audio. The allegation was direct: their voices appeared in publicly accessible catalogues, slightly altered but perfectly identifiable, without any authorisation having been given.
The legal qualification adopted by their lawyer is not traditional copyright law, but parasitic acts — a concept of unfair competition that allows action against the exploitation of someone else's work or reputation, even without intellectual property registration. The cease-and-desist letter set an eight-day deadline to remove the content, with a claim for damages for each affected actor.
VoiceDub removed some of the reported content within a few days. Fish Audio took longer, but eventually followed suit. In early April 2026, it was no longer eight but twenty-five French voice actors who had 47 vocal models relating to them removed from Fish Audio.
What this case shows, beyond the specific incident: a professional voice shared online — casting excerpt, demo, interview, session — can be captured and reproduced, often long before the artist concerned realises it.
Myths vs reality
False, but nuanced. There is not yet a specific law, but several existing legal bases already apply: neighbouring rights of the performing artist (article L.212-3 of the CPI), the right to privacy, and unfair competition through parasitic acts — the latter enabled takedowns in the VoiceDub/Fish Audio case.
True. The CNIL classifies the voice among elements that can identify a natural person, on the same level as an image. The GDPR therefore, in theory, governs its capture and reuse.
False. Vocal cloning catalogues operate on volume: the more a voice is available online, the more exploitable it is, celebrity or not.
True. You must demonstrate that the voice was actually used in the training data, or that the generated content reproduces its characteristics. A dated voice reference therefore has value in documenting this difficulty.
AI voice cloning, in practice
Vocal cloning relies on models trained to reproduce a voice's timbre, prosody and sometimes emotional inflections from an audio sample. Some platforms, such as Fish Audio in the case cited above, claim they can generate a credible clone from just ten to fifteen seconds of recording, with catalogues of several million voices.
For an artist, singer, voice actor, podcaster or content creator, a voice is already widely circulated long before any thought of cloning: casting extracts, demos sent to labels, interviews, live shows, stories. Every file becomes a potential source, often without the person concerned being informed.
To place this issue in the broader landscape of musical AI, also see our analysis of Suno and SACEM's position on AI — two related pieces that concern composition more than the voice itself.
What French law says today
Several legal texts are relevant, although none were written specifically for AI voice cloning:
Performing artists' neighbouring rights (Article L.212-3 of the French Intellectual Property Code): protects the fixation and reproduction of a vocal performance, which in principle includes a voice cloned from a real recording.
GDPR: The CNIL recognises the voice as personal data, which in theory governs its capture and reuse by a third party.
EU AI Act, Article 53: Applicable since August 2025, it imposes transparency obligations on AI providers regarding training data — a lever to require explanations.
Parasitic acts: A basis for unfair competition, successfully used in the VoiceDub/Fish Audio case, without requiring prior registration of intellectual property.
On the collective management organisations side, ADAMI — which represents performing artists, including dubbing and music voices — exercised its right of opposition (opt-out) as early as March 2025 for all of its 49,000 members, which presumes non-consent for any data mining of their performances.
And elsewhere in the world? The Tennessee case (ELVIS Act)
France is not the only territory facing a legal gap on this issue. In the United States, the state of Tennessee — the historic heart of the American music industry with Nashville — was the first to legislate specifically on AI voice cloning. TheELVIS Act (Ensuring Likeness Voice and Image Security Act), signed on 21 March 2024 and effective 1 July 2024, expands existing image rights to explicitly include the voice as a protected personal property right.
What the ELVIS Act makes possible, and which French law does not yet explicitly provide for: the law allows not only the user of an unauthorised voice clone to be sued, but also the publisher/developer of the technological tool itself, where its primary function is to create unauthorised vocal reproductions. Intentional infringements can give rise to treble damages.
This American precedent, combined with GEMA's July 2026 decision in Germany and the SACEM/ADAMI/ADAGP/SACD/Scam joint statement of May 2026, points to a clear trend: regulatory pressure on voice cloning is increasing across multiple continents at once, even though no harmonised framework yet exists.
Which protection covers what? Summary table
One of the most misunderstood points on this topic: these protections are not interchangeable; they are complementary.
SACEM
Protects: the musical composition (melody, lyrics, arrangement)
Does not protect: the performer's vocal identity
ADAMI opt-out (since March 2025)
Protects: presumes non-consent for the use of its 49,000 members' performances for AI training
Does not protect: non-member artists, and does not establish dated individual proof
Anti-AI contractual clause
Protects: governs how a client or label may use a specific recording
Does not protect: against a third party outside the contract who records the voice elsewhere
Timestamped voice reference (e.g. VoiceLockr)
Protects: establishes dated, verifiable proof of the existence of a reference voice at a given point in time
Does not protect: on its own it does not constitute a court decision nor a copyright registration — it is a piece of evidence, not a judgement
Key figures
🎙️ Create my Voice Passport on VoiceLockr
How to protect your voice: the method
1. Document a dated vocal reference before release
Before publishing a clip, sending a demo to casting or releasing a podcast, keeping a reference recording with a verifiable date greatly strengthens your ability to prove prior use of your voice in the event of a later dispute. This is the principle on which VoiceLockr : a time-stamped Voice Passport, without making the file public.
2. Check assignment clauses in contracts
A general clause authorising 'use' of a recording can, if poorly drafted, cover the training of an AI model. Requiring an explicit clause excluding any AI use without a separate written agreement has become recommended practice among several specialist law firms.
3. Limit exposure of long, isolated vocal excerpts
A clip several dozen seconds long, published without context, is more exploitable than a short fragment set against music or editing. It's no guarantee, but it reduces the surface area of exposure.
4. Monitor your own voice online
Periodically searching for your name alongside terms such as 'AI voice' or 'voice clone' can detect unauthorised use before it escalates — this is how several voice actors affected by the VoiceDub case initially spotted the issue.
5. Keep all evidence in case of a dispute
Original files, contracts, correspondence, screenshots of the disputed content: these items, together with proof of prior use, make up the basic dossier to hand to legal counsel if action becomes necessary.
Common mistakes
The first mistake is only worrying about the issue after a contested piece of content has been released — whereas the value of a voice reference lies precisely in its prior existence, created before any disputed use.
The second mistake is believing that only already-famous voices are affected. The VoiceDub/Fish Audio case shows that professional voices, even if they're not star voices, are equally exploited once enough material is available online.
The third mistake, more legal, is signing a voice services contract without reading the clause on assignment of usage rights. It is often there, rather than in outright theft, that implicit permission for an unanticipated AI use is hidden.
FAQ
Is AI voice cloning illegal in France?
There is no specific law, but several existing legal bases already allow action: neighbouring rights of the performing artist, the right to privacy, and unfair competition through parasitic acts — the latter enabled removals in the VoiceDub/Fish Audio case in 2026.
What should I do if I discover my voice has been cloned without authorisation?
Keep a capture of the disputed content, gather all evidence proving the prior existence of your voice, and contact a specialist lawyer or your collective management society (ADAMI). An amicable formal notice, as in the case of the French voice actors, can be enough to secure a swift takedown.
Does a simple SACEM registration protect my voice?
No. A SACEM registration protects a musical composition, not the performer’s vocal identity. These are two distinct but complementary protections.
How much audio is needed to clone a voice?
Some commercial platforms claim they can produce an exploitable clone from ten to fifteen seconds of recording, which makes even short clips potentially usable.
Am I protected if I’m not a professional artist?
The applicable legal foundations (neighbouring rights, privacy, the GDPR) are not limited to professionals. However, demonstrating economic harm is easier for someone whose voice is part of their professional activity.
🎙️ Protect my voice now
Additional resources
Timestamped voice reference. First Voice Passport free.
Protect the musical composition in addition to the voice.
The legal framework for musical composition.
The other side of musical AI: generation.
The complete guide to protection mechanisms.
Official sources cited:
- France Info — 25 voice actors obtain the removal of disputed content (April 2026)
- France 24 / AFP — VoiceDub withdraws its disputed content (February 2026)
- Digital Economy Blog — legal analysis of the case
- Haas Avocats — the GDPR and voice cloning, ADAMI opt-out
- Journal du Geek — context and accounts from the affected voice actors
- Davis Wright Tremaine — legal analysis of the ELVIS Act (Tennessee)
Last updated: 18 August 2026. This article will be updated as the VoiceDub/Fish Audio case evolves and a clearer legal framework emerges.
