Japan Recognises Voice as a Publicity Right in 2026

On 30 September 2026 the Tokyo District Court held, for the first time in Japan, that a person’s voice can fall within the scope of publicity rights. The claim in front of it still failed. Seven weeks earlier the Ministry of Justice had reached the same conclusion in a 143-page interpretation report, and on 15 October 2026 the union representing Japanese voice actors published its own consent rules for AI.

Japan has no statute that names a right in the voice. What it now has instead are three dated documents that say, in different registers, that the voice is protected anyway. This page sets out what each one actually decided, where the gaps still are, and what it changes for anyone commissioning Japanese-language voice work.

The ruling: a voice actor, TikTok, and a case that was already over

The actor and voice performer Kenjiro Tsuda sued the operator of TikTok to have videos removed that used audio imitating his voice. The Tokyo District Court ruled on 30 September 2026. It dismissed his removal claim, because the account and the videos had already been deleted before judgment, which left nothing to order.

The reasoning is the part being cited. According to reporting by Jiji Press and Nikkei, the court held that a human voice is included in the subject matter of publicity rights, treating it as a symbol of individual personality, and that infringement can arise where a voice carries power to attract custom and is used for the sole purpose of exploiting that power.

The Japan Actors Union statement of 5 October 2026 frames the outcome as a win on both counts: the videos came down, and the judiciary said what the union had been arguing. Two cautions belong on any citation. This is a first-instance district court judgment, not a Supreme Court precedent, and the voice point was reasoning rather than the operative order. And it concerns a performer whose voice is extremely well known in Japan — the customer-attracting power the court spoke of is not a given for everyone.

What the Ministry of Justice report says

The more systematic document came first. The Ministry of Justice convened a study group on civil liability for the unauthorised use of likeness and voice, chaired by Professor Yoshiyuki Tamura of the University of Tokyo. It met five times between 24 April and 27 July 2026, and its final report, “Interpretation guidance on publicity right infringement by generative AI”, is published on the ministry’s own page as a full PDF with a separate summary. It is interpretation of existing law, not new law, and it binds nobody.

Four strands in it matter for voice work.

Publicity rights. The report treats a voice as identifying information that functions as a symbol of personality, and therefore as protected subject matter alongside name and likeness. Infringement is framed around three typical patterns: the voice used as the product itself, the voice attached to goods to differentiate them, and the voice used in advertising.

The separate personality strand. Independently of commercial value, the report recognises a right not to have one’s likeness or voice used arbitrarily, protecting non-economic interests. The test is whether the interference exceeds what has to be tolerated in social life, weighed against the person’s position, the purpose and the manner of use. This is the route that does not require fame.

Unfair competition. Where a well-known person’s voice functions as an indication of goods or business, the Unfair Competition Prevention Act may apply, which also opens standing to agencies rather than only the performer.

After death. The report records that the inheritability of publicity rights remains unsettled in Japan, that general personality rights are understood to end with the person, and that unauthorised use of a deceased performer’s voice may still be actionable through the surviving family’s interest in honouring the dead.

On generative AI specifically, the report makes a point that anyone building a takedown case should read twice: because AI output is rarely identical to its source, identifiability may turn on context — titles, tags, framing — as much as on how similar the audio sounds.

Why Japanese performers say copyright does not help them

The union’s reading of its own legal position is blunter than the ministry’s. In its statement on the TikTok ruling it points at three features of the Japanese Copyright Act, in force since 1970, that leave voice performers without a straightforward basis even to demand that unauthorised use stop.

• Article 30-4 permits use of works where the purpose is not to enjoy the expression itself — the provision generally relied on for machine learning on existing material.
• Article 91(2) means a performer’s recording rights do not extend to cinematographic works, which is most dubbing.
• Performers hold no general reproduction right.

That is the gap the other two documents are working around. European readers will recognise the shape of the first point: the EU text and data mining exception has the same structure, except that it comes with an opt-out, which Japan’s provision does not. We cover that mechanism and its limits in our page on the AI training opt-out.

The Japan Actors Union guidelines of 15 October 2026

The Japan Actors Union, a cooperative with roughly 2,600 actor and voice-actor members, announced guidelines on the use of generative AI for voice at a press conference on 15 October 2026, with the full text published as a PDF. The union states its reason for acting plainly: there is no comprehensive statutory voice right in Japan, and waiting for legislation would be too late. The guidelines are self-regulatory. They are addressed as much to production companies, platforms and AI developers as to performers.

Four requirements carry the document.

1. Consent at every stage, not once. Training (input), generation, and publication or use (output) are treated as three separate permissions. The purpose must be stated and consent obtained in advance for each.
2. Payment as performance. AI-generated use of a voice is defined as the equivalent of a new form of providing a performance, and is to be paid accordingly, scaled to the scope and record of use.
3. Multilingual output is a new licence. Automatically translating a performance into other languages requires fresh permission and additional compensation, with the stated aim of protecting the artistic identity of the performance and preventing use in territories outside the contract.
4. The dead are not a free archive. Reviving a deceased performer’s voice requires strict permission from the heirs or estate, and the union asks that it not be done for new works as a matter of principle.

Alongside the guidelines, the union has asked operators not to substitute AI for human performers in film and drama dubbing, and says it will press the government for statutory recognition of a voice right. Reporting around the launch put a well-known face to the argument: Masako Nozawa, the voice of Goku, spoke publicly about unauthorised AI use of her voice on the day of the announcement.

The identification problem, and some numbers

Every voice-cloning dispute eventually becomes an evidence problem: can you show the audio is you? The union runs a performer database, JVOXPRO, with a similarity-scoring tool, and it published both its thresholds and its results in the Tsuda matter.

The bands it uses are 0.90 and above for “very likely the person”, 0.80 to 0.90 for “likely the person or AI-generated audio of the person”, 0.74 to 0.80 as a zone requiring checking, and below 0.74 for “likely someone else”. Across 30 test comparisons between Tsuda’s voice and the audio used by the poster, the union reports a maximum score of 0.881, more than fifteen results at or above 0.80, and an average of 0.814.

Treat those figures as what they are: a union’s own tool, reported by the union, with no court endorsement of the methodology. The union itself concedes the harder point — identification short of certainty leaves room to litigate, and that risk alone deters claims.

What changes if you commission Japanese voice

Nothing in Japan has become illegal that was legal in August. What has changed is that three authorities now point the same way, which makes a buyout clause that silently covers AI use harder to defend and easier to challenge.

Practical consequences for buyers, in order of how much trouble they save:

• Split the permissions. If you want training rights, generation rights or synthetic redubbing, ask for each separately and price each separately. The union’s three-stage model is now the reference point a Japanese agency will measure your contract against.
• Treat AI localisation as a new engagement. Running a Japanese performance through a multilingual voice model is, on this reading, a fresh use needing fresh consent — not a delivery format.
• Document the chain. Where a vendor supplies synthetic Japanese audio, ask whose voice the model was built on and what that person agreed to, in writing.

The pattern is familiar from Europe, where the argument has been running through contracts rather than courts. German dubbing has been in open conflict over an AI training clause since early 2026, which we track on our Netflix dubbing page, and the EU has layered a labelling duty on top of the consent question in Article 50 of the AI Act. For how the underlying right differs by jurisdiction, see who owns your voice, country by country.

And the simplest route is still available. If you need a Japanese, German or Spanish voice for a campaign and you would rather not own a consent problem, briefing a native performer removes the question instead of documenting it.