AI Act: what changes for the voice‑over industry

In recent years, voice synthesis technologies have made huge progress, moving from something experimental to tools used every day in areas like e‑learning, social content and even dubbing. This very fast growth has brought enthusiasm and curiosity, but also a certain amount of confusion about what is allowed, what the ethical limits are, and how the human voice can be protected. With the European AI Act coming into force on August 2nd, 2026, this scenario finally starts to have a clearer and more protected framework.

The AI Act introduces a series of obligations that directly concern synthetic voices and voice‑cloning systems - and not only them, since the regulation also covers many other fields where artificial intelligence is used, such as the medical and healthcare sector. Here, however, I will focus on the area I work in: the voice.

Every piece of content generated through AI on dedicated platforms will need to be clearly recognisable through invisible markers inside the audio track and through explicit written labels, so that an artificial voice cannot be mistaken for a real one. By “label”, the regulation means both visible notices for the user and machine‑readable metadata in standardised formats, which allow automatic tracking of the content. This does not mean adding audible watermarks or elements that disturb the listening experience. It means adopting technologies able to identify the artificial origin of the audio even after modifications, compression or editing. These markers must be resistant to this type of manipulation.

This is also a protection for the listener, who will finally know the true origin of the content they are interacting with.

Another important point is that the AI Act does not apply only to European companies, but to anyone publishing content accessible within the European Union. This means that an AI‑generated video, a podcast or a dub created in other continents will still need to respect these rules if they reach a European audience. Providers of synthetic and cloned voices, platforms that distribute them, and those who publish the content all have different but connected responsibilities. International platforms like YouTube, TikTok or Spotify will be the first to require compliance, making transparency a global requirement.

For voice‑over professionals, this new regulatory framework represents an important protection and gives all of us a bit of relief. A voice cannot be used to train AI models without explicit, informed and documented consent. It is important to remember that consent is part of a wider framework that also includes GDPR. Companies recruiting voices for synthesis or cloning projects must explain how the recordings will be used, for how long, in which territories and with which limitations. They must also guarantee the possibility to withdraw consent, with documented procedures and clear timelines. Withdrawal obliges companies to stop future uses and delete the data where possible, even if it is not always technically easy to remove information already incorporated into a model without specific interventions. Companies must protect voice data from unauthorised use and document, in a traceable way, the entire life cycle of AI‑generated content. This means keeping exportable control registers and defining clear retention and deletion policies.

The human voice will regain a recognisable and protected value, not only from an artistic point of view, but also from a legal and technological one.

In my work, the arrival of the AI Act represents an opportunity for clearer and more responsible collaborations. When a company proposes a project involving voice‑synthesis technologies, the voice talent will finally have the law on their side, with the right to ask precise information about the intended use, the management of voice data and the transparency offered to final users. This is not about distrust, but about ensuring that every collaboration is built on solid, respectful and compliant foundations.

Artificial intelligence is a powerful tool that can expand creative possibilities and simplify many processes. But the human voice remains an experience, a gesture, a sound that cannot be fully replicated, because it is unique and supported by the experience and the personal story of the person interpreting the text.

Choosing a human voice means choosing authenticity, relationship, interaction and care. It is a choice that enriches the project and respects the listener.

For those who want more precise regulatory references, the AI Act details the obligations of transparency and marking (especially the rules on disclosure and technical requirements for AI‑generated content), and new guidelines and a Code of Conduct are being developed to standardise labels and metadata. To know more click here