This is a clear setback for Fish Audio, and a strong signal for the entire industry: 25 French voice actors have obtained the removal of 47 generative artificial intelligence models that were reproducing their voices without their consent or compensation, according to their lawyer, Jonathan Elkaim. For the artists, this removal marks a first concrete victory in a battle that has become emblematic of the abuses of voice cloning.
From a protest by eight artists to a collective response
The case didn't arise overnight. In early February, eight voice actors launched a legal offensive against Fish Audio and VoiceDub after receiving cease-and-desist letters denouncing "parasitic practices." In letters dated January 30, they demanded the removal, within eight days, of the models using their voices and claimed €20,000 in damages. Since then, the group has grown: twenty-five voice actors have now joined this action, which is being described as unprecedented in the industry.
Voices transformed into a commercial catalog
At the heart of the dispute lies a very concrete use of these tools: for a fee, users can have any text read aloud by a voice chosen from a catalog. For voice actors, this is simply theft! Their voice is both a tool of their trade and an artistic identity, not to mention the performances they have honed over the years. The content in question featured artists well-known to French-speaking audiences, including voices associated with Julia Roberts, Richard Gere, Angelina Jolie, Buzz Lightyear, and Cartman.
Familiar names
Among the first plaintiffs were Céline Monsarrat, Delphine Allemane, Philippe Ariotti, Adrien Antoine, Françoise Cadol, Richard Darbois, Brigitte Lecordier, and Christophe Lemoine. As early as February, several of them publicly denounced what they called a blatant misappropriation of their work. On TF1, Céline Monsarrat, the French voice of Julia Roberts, summed up the industry's shock in three words: "It's theft." Behind this statement lay the same fear among them all: seeing their voices captured, imitated, monetized, and then used without any oversight.
After VoiceDub, Fish Audio also retreats.
A first breach had already occurred in February when VoiceDub immediately removed the content targeted by the initial cease-and-desist letter. The artists' lawyer mentioned the removal of about ten illicit voices and performances from the platform. Two months later, Fish Audio followed suit, removing 47 tracks. The balance of power has clearly shifted: what seemed like an isolated protest has now become an organized and effective response.
A symbolic victory, but not yet a legal one.
However, the voice actors are far from declaring a definitive victory. Jonathan Elkaim says he remains "wary" : the compensation claims, particularly the €20,000 demanded, have not been met, and there is no guarantee that new contentious material will reappear. The lawyer explains that he is even working on a new legal action aimed, ultimately, at obtaining a landmark court ruling, or even banning these platforms from French territory. In short, the removal of the content has calmed the conflict, but it does not resolve the dispute.
French dubbing enters into resistance
The #TouchePasMaVF collective has been advocating since 2024 for "dubbing created by humans for humans," a phrase echoed by the association LesVoix. Its petition has now surpassed 255,000 verified signatures on Change.org. At the end of February, the protest escalated further: 4,000 actors, actresses, and filmmakers signed an open letter denouncing the "outright plundering" made possible by AI tools capable of reproducing voices and images without authorization.
Law is finally catching up with technology
The plaintiffs' argument is simple: because these services are accessible in France and marketed to a French audience, they cannot be exempt from the protections afforded by French law. The voice actors' lawyer argues that voice-over work falls under the rights of personality and privacy, thus allowing them to sue platforms based in the United States. At the European level, the regulatory timeline is also tightening: the Commission reiterates that the AI Act's transparency rules on AI-generated content will come into effect on August 2, 2026, with identification and labeling requirements, particularly for certain synthetic content and deepfakes. However, this case implicitly reveals the limitations of transparency alone: labeling content does not address either the issue of consent or that of remuneration.
