Nothing in European or American law stops a brand from broadcasting a synthetic voice. That part is settled, which is why it is almost never the question that costs anything. The expensive one arrives later, usually during a legal review of a campaign that is already on air: where did this voice come from, and who answers if a person recognises themselves in it? The texts have moved a great deal in two years, in Brussels and in several American states. Here is what they say, checked at source.<\/p>\n\n\n\n
There is no single voice statute on either side of the Atlantic. What protects a voice is a stack of separate rules, and the stack is not built the same way in Europe and in the United States.<\/p>\n\n\n\n
In France, where our booth is, the personality layer became explicit in 2026. On 24 June, the first civil chamber of the Cour de cassation held, in a ruling published au bulletin<\/a> (pourvoi n\u00b0 25-20.483), that like their image, a person’s voice is one of the principal attributes of their personality. It is protected on the same footing, under article 9 of the Code civil and article 8 of the European Convention on Human Rights.<\/p>\n\n\n\n Protection is not a general prohibition. The same ruling requires a judge faced with a conflict between artistic expression and the voice right to weigh the two and let the more legitimate one prevail, taking into account the contribution to a debate of general interest, the notoriety of the person, and the content and effects of the publication. The practical reading for an advertiser is easy to miss: the factors that can tip the balance the other way are public debate and creation. A commercial message sits comfortably in neither.<\/p>\n\n\n\n The second layer is the one buyers forget at the moment of purchase. Article L. 212-3 of the Code de la propri\u00e9t\u00e9 intellectuelle<\/a> makes the fixing of a performance, its reproduction and its communication to the public subject to the written authorization of the performer. Two bodies of rules, so two separate consents: the person’s, over an attribute of their personality, and the performer’s, over their work.<\/p>\n\n\n\n The United States starts somewhere else, from the right of publicity, which is state law and differs from one state to the next. Tennessee rewrote its version in 2024. The Ensuring Likeness, Voice, and Image Security Act<\/a>, Public Chapter 588, effective 1 July 2024, defines a voice as “a sound in a medium that is readily identifiable and attributable to a particular individual, regardless of whether the sound contains the actual voice or a simulation of the voice of the individual”. Read the last clause twice. A simulation is treated exactly like the real thing, and nothing turns on how the sound was produced. Tennessee is one state among fifty, so the states where your campaign will actually run are a question for your brief, not for your legal team six months later.<\/p>\n\n\n\n Recording a voice and training a model on that voice are two different operations, and they call for two different authorizations. A session signed in 2019 carries no training permission, and no broadly worded assignment clause fills that gap after the event.<\/p>\n\n\n\n California has written into statute how specific the paperwork has to be. Labor Code section 927<\/a>, added by AB 2602 and in force since 1 January 2025, makes a contract provision permitting a digital replica of a performer unenforceable where it lacks a reasonably specific description of the intended uses of the replica and the individual was not represented either by counsel who negotiated the terms or by a union whose agreement expressly addresses digital replicas. It bites on replica uses fixed on or after 1 January 2025. A signature at the bottom of a wide clause does not, on its own, amount to consent.<\/p>\n\n\n\n Three questions decide whether an AI voice cloning file holds up under challenge. Form: is the performer’s authorization in writing, and does it name the training of a model rather than leaving it implied? Scope: which uses, which media, which territories, for how long, and what happens when the contract ends while the model carries on existing? Proof: who holds the signed document, and can you produce it the same day someone contests it?<\/p>\n\n\n\n That third question is where most files come apart. Not because nobody consented, but because nobody can show it.<\/p>\n\n\n\n From the booth<\/strong><\/p> Consent never comes up during a session. It surfaces months later, on the day the campaign works: the spot is held over for a second flight, a neighboring market picks it up, someone cuts a fifteen-second version for social. At that point the scope agreed before the recording is the only document in the room. That is why we ask for the intended uses before casting rather than after the mix, because a performer commits their voice to that scope and to nothing beyond it.<\/p><\/blockquote>\n\n\n\n Buy an AI voice over off the shelf and the training data stays opaque to you. You get a timbre. Where it came from is not part of the delivery.<\/p>\n\n\n\n The risk does not sit in the quality of the render, and this is the part that surprises marketing teams. It triggers on the day an identifiable person believes they can hear themselves, and that person does not go after the model. They go after what they heard, which is your campaign.<\/p>\n\n\n\n The Tennessee text is blunt about who that reaches. A person is liable in a civil action if they publish, perform, distribute, transmit or otherwise make available to the public an individual’s voice, knowing the use was not authorized by that individual. A separate limb reaches whoever distributes an algorithm, software or tool whose primary purpose is producing a particular identifiable individual’s voice. Both ends of the chain are exposed. Only one of them has your campaign’s name on it.<\/p>\n\n\n\n There is a test for this that requires no legal training. Open your vendor’s terms and find the warranty and indemnity clause: it will tell you, more reliably than any product page, who carries the risk and up to what ceiling. Then put the question the other way round, the way legal departments increasingly do in writing: can the files we hand you be used to train a model? With no explicit clause, your brand voice becomes raw material for a third party.<\/p>\n\n\n\n Regulation (EU) 2024\/1689<\/a>, the AI Act, handles the question through transparency, and it splits the duty between two parties.<\/p>\n\n\n\n Article 50(2) puts marking on the provider: anyone supplying an AI system generating synthetic audio must ensure the outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated”. Article 50(4) turns to the deployer, meaning whoever uses the system under their own authority, which is the advertiser or the agency: where audio content constitutes a deep fake, the deployer has to disclose that it was artificially generated or manipulated. Article 50(5) adds that the information must reach people “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”.<\/p>\n\n\n\n Two dates settle the calendar, and both are checkable. Article 113 sets general application at 2 August 2026, and the chapter containing article 50 is not among the listed exceptions, which the European Commission’s own page<\/a> confirms for the transparency rules. Regulation (EU) 2026\/1744 of 8 July 2026<\/a>, the Digital Omnibus on AI, published in the Official Journal on 24 July 2026, moved the high-risk obligations to 2 December 2027 and 2 August 2028 without touching transparency. It did add one breathing space that matters here: a new article 111(4) gives providers of generative systems already placed on the market before 2 August 2026 until 2 December 2026 to comply with the article 50(2) marking duty.<\/p>\n\n\n\n Two uncertainties, which we would rather flag than resolve for you. The definition of a deep fake, at article 3(60), covers content that “resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”. Whether a wholly synthetic voice imitating no one in particular falls inside that wording is still argued, and there is no settled decisional practice to point to. The concrete form of the disclosure is also fixed by no text today: article 50(7) sends that question to codes of practice still being drawn up, and the July 2026 omnibus rewrote the paragraph.<\/p>\n\n\n\n One consequence catches radio campaigns in particular. Article 50(4) narrows the duty for content forming part of an “evidently artistic, creative, satirical, fictional or analogous work”, where the disclosure must simply not spoil enjoyment of the work. A brand film does not walk into that category on its own. On air, with no screen and thirty seconds to work with, a disclosure that has to land at first exposure belongs in the script at brief stage. Bolted on at the mix, it eats airtime and unbalances the writing.<\/p>\n\n\n\n Nobody asks this before signing, and it decides everything else.<\/p>\n\n\n\n The AI Act allocates the roles itself: machine-readable marking on the provider, public disclosure on the deployer. The Tennessee statute attaches liability to publication and distribution rather than to generation. A civil claim in France goes after the visible interference, which is the broadcast rather than the code that produced it. One asset running in two markets therefore answers to two regimes at once, each following the place where the campaign is seen.<\/p>\n\n\n\n Your vendor’s contractual warranty is a genuine remedy, and we would never tell a client to ignore it. It runs between you and your vendor. It never runs between you and the third party who recognises themselves, it is worth exactly what your counterparty’s solvency is worth, and it does nothing about the film still playing while the dispute runs its course. Responsibility for what a brand broadcasts is not subcontracted.<\/p>\n\n\n\n Three requirements, and they apply to a human voice exactly as they apply to an AI voice over.<\/p>\n\n\n\n Traceable origin: who spoke, when the performance was fixed, on what basis. The industry has an open technical standard for this, C2PA<\/a>, which attaches signed provenance metadata to a file.<\/p>\n\n\n\n Documented consent: written, dated, with a scope that names the uses, the territories, the duration, and the position taken on model training.<\/p>\n\n\n\n A written chain of rights running from the performer to the advertiser, with no gap between two intermediaries. The whole thing fits in one email: ask your supplier where the voice in your last campaign came from. How long the answer takes will tell you more than any commercial promise.<\/p>\n\n\n\n For 15 years, our voice over agency<\/a> has been building those chains of rights for campaigns running on television, on radio and online, before the casting conversation even opens. If you are preparing a corporate campaign<\/a> and you are unsure what your scope actually covers, tell us about the project<\/a> and we will tell you what the file is missing.<\/p>\n","protected":false},"excerpt":{"rendered":" Nothing in European or American law stops a brand from broadcasting a synthetic voice. That part is settled, which is why it is almost never the question that costs anything. The expensive one arrives later, usually during a legal review of a campaign that is already on air: where did this voice come from, and…<\/p>\n","protected":false},"author":1,"featured_media":21689,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[99],"tags":[],"class_list":["post-22646","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-rights-and-safety"],"_links":{"self":[{"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/posts\/22646","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/comments?post=22646"}],"version-history":[{"count":0,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/posts\/22646\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/media\/21689"}],"wp:attachment":[{"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/media?parent=22646"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/categories?post=22646"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/frvoiceover.com\/en\/wp-json\/wp\/v2\/tags?post=22646"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Cloning an existing voice: what the consent has to cover<\/h2>\n\n\n\n
The blind spot: where the voice you broadcast came from<\/h2>\n\n\n\n
What the European rule requires, and from when<\/h2>\n\n\n\n
Who answers a claim: the tool vendor, or the brand that published<\/h2>\n\n\n\n
What to require, whichever technology you choose<\/h2>\n\n\n\n