What just happened in Tokyo
On October 15, 2026, the Japan Actors Union, which represents roughly 2,600 actors and voice actors, held a press conference to release the first formal guidelines in Japan governing how generative AI can be used on vocal performances. The document is 21 pages long, split into six sections and 22 articles, titled Guidelines for the Use of Generative AI on Vocal Audio Work.
The timing wasn't an accident. The announcement landed alongside a public statement from Masako Nozawa, the 89-year-old voice actress who has played Goku in Dragon Ball for over four decades. Nozawa said AI clones of her Goku performances were circulating without her permission, and argued that synthetic voices lack the emotion real actors bring to a role. That combination, a union policy document plus a beloved voice actress calling out unauthorized clones, pushed the story into mainstream coverage through late October.
The rules in plain terms
The guidelines center on a simple idea: consent has to happen at every stage, not just once. That means a company needs permission before training an AI model on an actor's voice, before generating synthetic audio from that model, and before distributing anything made with it. A contract that covered one use doesn't automatically cover the next one, and continued use after a contract expires requires fresh consent.
If an AI system replicates a performance, the union says that output should be treated as a new performance in its own right, which means the actor is owed additional royalties or compensation, not just a one-time licensing fee paid upfront. The guidelines also address AI-powered dubbing and translation: automated translations into another language must preserve the artistic integrity of the original performance, and they can't be used outside Japan without separate permission.
- Consent required before training, before generating, and before distributing AI voice output
- AI-replicated performances treated as new performances requiring additional pay
- Automatic translations must preserve artistic integrity and need separate permission for overseas use
- Continued use past a contract's end date requires renewed consent
The hard line on deceased actors
The clearest, least negotiable part of the guidelines concerns actors who have died. Under the new framework, a deceased actor's voice can only be recreated with AI if that actor gave permission while they were alive. Even then, developers also need consent from the actor's family and their affiliated talent agency, plus they have to pay royalties.
This directly targets a practice that's become common in faceless and parody content: cloning the voice of a dead celebrity, narrator, or public figure because they can't object and there's no clear legal owner of their voice to license from. JAU's position closes that loophole, at least as a matter of industry norms in Japan. There's no indication in the guidelines of any exception for parody, satire, or non-commercial use.
Why this matters even though it's not law
It's worth being precise about what these guidelines actually are: voluntary. Japan has no law that treats a person's voice as something they legally own and control, and the union says so directly in its own document. JAU is hoping the industry adopts these guidelines on its own, and that lawmakers eventually follow with actual legislation establishing what the union calls voice rights, enforceable legal protections covering both an actor's natural voice and any AI-generated version of it.
There's already a legal nudge in that direction. A Tokyo District Court ruling has recognized that unauthorized voice cloning can infringe on a performer's publicity rights, giving the union's push some real legal grounding to build on. And this isn't happening in isolation. After an 11-month strike, SAG-AFTRA members in the US voted in favor of the 2025 Interactive Media Video Game Agreement, which included its own consent and disclosure requirements for AI digital replica use in games. Two separate unions, two separate countries, converging on the same basic framework: ask first, pay for new uses, disclose when AI is involved.
The brief is clear that none of this is enforceable law yet, in Japan or elsewhere. But guidelines like this tend to spread. Platforms, licensing marketplaces, and other countries' performer unions often look at documents like JAU's as a template, especially once a court ruling and a strike settlement are already pointing the same direction.
What this means if you use AI voices in your content
If you run a faceless channel, voice a recurring AI character, or use voice cloning tools to narrate videos, none of JAU's rules apply to you directly unless you're working with Japanese talent or distributing in Japan. But the direction of travel matters more than the jurisdiction. Consent-and-compensation norms that start with unions and court rulings have a track record of hardening into platform policy, and eventually law, well beyond where they began.
The practical risk is narrower than it sounds: cloning a real, identifiable actor's or celebrity's voice without a license, and cloning the voice of someone who's died, regardless of whether a specific law currently bans it where you live. Both practices are exactly what JAU's guidelines target, and both are common shortcuts in parody and reaction content because they're free and nobody's been enforcing against them yet.
The more durable move is to shift toward voices you actually have rights to: licensed synthetic voice libraries with clear commercial terms, or an original AI voice persona you build and control yourself rather than a clone of someone real. Tools like Episodo that let you generate distinct, ownable voice personas for recurring characters sidestep this problem entirely, since you're not relying on anyone else's likeness or performance. That's a cleaner position to be in now, before consent requirements that are voluntary in Japan today become platform-enforced or legally required somewhere you publish tomorrow.