Japan Actors Union Seeks Consent, Royalties and Limits on AI Voice Replicas

Permission before training, payment when a replica performs

The Japan Actors Union, a cooperative representing about 2,600 actors and voice actors, has set out rules requiring permission before artificial intelligence learns, generates or publishes a replica of a performer's voice. Its guidelines also call for payment when generated speech is used, fresh approval for each language in AI dubbing, and strict limits on recreating the voices of deceased actors.

The union dates the formal announcement to October 15, 2026, although the announcement has also been described as taking place on October 16. The standards reach beyond permission to make a recording: they seek to give performers control over how a digital version of their voice is built, what it says and where it is heard.

In its voice AI guidelines announcement, the union identifies a central problem: Japan has no comprehensive law specifically protecting an individual's voice. It wants the standards adopted across the industry while pressing the government to legislate.

For performers, the practical concern is straightforward. A recording made for one role can become material for a system that produces words or songs the actor never performed. For production companies and AI developers, the guidelines propose a process for securing permission and budgeting for that use.

The standards are voluntary industry guidance rather than a new national law. Their reach will depend on adoption by companies, their inclusion in contracts and future legal changes. The Japan Entertainments Management Entrepreneurs' Association and Japan Voice Actors Association reportedly support them, giving the proposals backing beyond the union itself.

Consent must cover three separate stages

Part 3, Articles 7 through 9, divides AI voice use into training, generation and publication or use. The union says the purpose must be explained and advance permission obtained at every stage.

Training means feeding recordings into a system so it can learn patterns in a voice, such as pitch, rhythm and pronunciation. Generation is the creation of new audio using that system. Publication or use is the point at which the resulting speech reaches an audience or becomes part of a production.

A company following the standards would need to address all three stages. Permission to train a model would have to be followed by permission for generation and the intended release. Additional generated content would also require consent.

The union says this control matters because a synthetic voice can place an actor in a role they never accepted or make them appear to express political or violent statements they never chose to deliver. Disclosing the intended purpose gives performers a chance to decide before the audio exists.

Performers who refuse permission, or later withdraw it, must not lose casting opportunities, receive worse employment conditions or face other disadvantages under the guidelines. That protection addresses the pressure an actor could face when deciding whether to license a voice to a potential employer.

A generated voice would earn compensation

Part 5, Article 16, treats the use of generated audio as a new form of providing a performance. The union's position is that actors should receive suitable compensation, including royalties, according to the extent and manner of use.

A royalty is a payment tied to the use of someone's work or licensed rights. Here, the proposal would let a performer earn money when an authorized replica is used, rather than leaving the payment question settled by the original recording session alone.

The union presents licensing as a possible source of revenue for actors and a way for companies to plan production costs. The announcement gives no universal royalty rate or payment formula. Those terms would need to be settled through the relevant agreements.

Actors describe voices used without their knowledge

Bin Shimada, the union's managing director and the voice of Tomozo in the anime Chibi Maruko-chan, said his own voice had been used without permission. At the announcement, he described hearing an AI cover of Ado's song Usseewa using the voice of another character he performs.

Shimada explained how his reaction changed after hearing the recording:

The first time I heard it, I was stunned. As time passed, my anger grew. Words I never said and songs I never sang were created without permission, uploaded to the internet and monetized. No matter how I look at it, I simply cannot accept that.

Masako Nozawa, a union vice chair known for voicing Son Goku in Dragon Ball and Hoshino Tetsuro in Galaxy Express 999, has also publicly challenged unauthorized replicas. She described her voice and performances as her own and said permission should be required even for partial use.

Nozawa said an AI version she heard sounded similar to her voice but lacked the feeling she brings to a performance. Her objection concerns both control over her identity and the acting choices behind the sound.

Some performers have supported private systems that replicate voices under licence. The union's standards allow a route for authorized use while putting human performers first.

AI dubbing faces rules in both directions

Part 4, Article 15, addresses the use of AI to turn an original performance into speech in another language. The union says the artistic identity of the performance must be preserved, with new permission and additional compensation for wider international use.

Approval would be needed for each language. A company authorized to produce an English version, for example, would need further approval before generating a Chinese or Spanish version. Performers would also retain control over use in overseas territories outside their agreements.

AI localization means adapting a production for audiences in another language or market. A voice replica can make a character sound like the original performer while speaking translated words. The union's concern is that changes in wording, timing or tone can alter the acting as well as the language.

The standards also address the other direction: dubbing foreign productions into Japanese. Replacing Japanese voice performances with AI is generally prohibited under the guidance. Exceptions require permission from performers, rights holders and audio producers, supported by individual written contracts specifying compensation and other terms.

Deceased actors would receive special protection

Part 2, Article 6, sets out protections for performers who have died. The standards generally rule out using AI to recreate their voices, with limited exceptions for projects the performer agreed to during their lifetime.

Even those projects would require written permission from the bereaved family and the actor's former agency or affiliated organization, along with suitable compensation. The union says the performer's wishes, dignity, reputation and legacy must guide decisions.

For an archive or entertainment project, this would make authorization a process involving several parties. A producer seeking to revive a familiar voice would need to establish the actor's prior agreement and secure the required approvals before proceeding.

The restrictions matter because an actor can no longer review a proposed script, reject a role or explain how they want a new performance delivered after death. The union seeks to place those decisions within the boundaries the performer accepted while alive.

A proposed committee would review complete replacement

Projects that entirely replace human performances with AI voices would face another step: prior consultation with, and written approval from, a proposed industry body tentatively called the Voice AI transaction fairness committee.

The committee's structure and establishment date have not been settled. Its proposed role places the decision to remove human performances from an entire project under industry review, alongside the permissions required from individual performers.

The actors' demands also include destruction of unauthorized AI voice models. That reaches beyond removing a particular clip: a model can remain available to produce more audio after an offending recording disappears.

For violations, the union says it will demand an immediate halt to distribution and sales of the affected content. It would also seek payment equivalent to sales dating back to the start of unauthorized use. These are the remedies the union intends to pursue; their application would depend on the relevant agreements and legal proceedings.

The government is considering protection for voices

Japan's industry ministry is considering whether voices and likenesses should become protected assets under the Unfair Competition Prevention Act. The law already provides a route to seek injunctions against conduct that distorts fair competition. The ministry plans to discuss the issue with an expert panel.

Publicity rights are another part of the debate. They concern a public figure's control over the commercial use of their identity. For a voice actor whose recognizable sound helps sell entertainment, an AI imitation can turn that identity into a commercial product without their involvement.

Voice actor Kenjiro Tsuda's legal action over imitation videos on TikTok has brought the issue into focus. The Tokyo District Court was reported to have recognized that a voice can receive protection under publicity rights. The lawsuit was also reported as dismissed after the videos were removed; the precise relationship between that finding and the dismissal is unclear.

The union's proposals extend into matters the government has yet to resolve, including unauthorized models, performers who withdraw consent and synthetic performances after death. No legislative vote, implementation date or date for the proposed expert discussions has been announced.

How the campaign reached this point

Japanese performers have been organizing against unauthorized AI voice use for several years. The new standards bring that campaign into a more detailed framework covering contracts, payments and production decisions.

• In 2023, SAG-AFTRA members authorized a strike against major video game publishers while seeking protections concerning AI and job security.
• In 2024, Japanese voice actors formed an organization protesting misuse of their voices by AI and gaps in legal protection.
• In 2025, after an 11 month strike, SAG-AFTRA members approved an Interactive Media Video Game Agreement containing AI safeguards.
• On October 15, 2026, according to the Japan Actors Union's announcement, the union formally presented its voice AI guidelines.

The American video game agreement requires consent and disclosure for digital replica use. It also allows performers to suspend consent for the generation of new material during a strike. Japan's proposals cover similar concerns about permission while adding detailed demands on dubbing, deceased performers and projects that replace human acting completely.

AI adoption makes those questions immediate for the games business. Research cited in the original announcement coverage found that more than half of Japanese developers used generative AI in game development, chiefly for visual assets, character art, stories, game text and programming support. Voice replication would bring another group of creative workers into decisions about how those tools are used.

Human performances remain the priority

Michihiro Ikemizu, the union's vice chairman, described the aim as creating conditions in which AI can be used with confidence:

We need rules that enable AI to be used properly, safely and with peace of mind.

Nozawa also expressed support for a future that includes the technology. In a video message for the union's announcement, she urged people to follow the rules:

I hope everyone follows the rules so that both people and AI can enjoy a future where they can happily coexist.

The union's approach combines that openness with a clear priority for employing human performers. Its proposed bargain is permission before use, payment for authorized replicas and limits where a performer cannot meaningfully take part in the decision.

For actors, the immediate change would be greater control in contracts. For producers, developers and distributors, it would mean checking rights at each stage rather than treating a recording as permission for every later AI use. Turning those expectations into a consistent industry practice is the union's next task, alongside its campaign for legislation.

Key Points

• The Japan Actors Union represents about 2,600 actors and voice actors and dates its guidelines announcement to October 15, 2026.
• The standards require advance consent for training, generation and publication or use of AI voice replicas.
• Authorized generated performances should earn compensation, with terms reflecting how extensively the voice is used.
• AI dubbing requires approval for each language, while replacing Japanese dubbing performances with AI is generally prohibited.
• Recreating a deceased actor's voice would be limited to previously agreed projects with further written permissions and payment.
• A proposed industry committee would review projects that completely replace human performances.
• The government is considering legal protection for voices and likenesses, but no legislative timetable has been announced.