A Tokyo court has delivered a consequential ruling for performers in the age of synthetic media: an actor’s voice can be protected from unauthorized AI imitation. The case involved Kenjiro Tsuda, the prolific Japanese voice actor widely associated with Seto Kaiba in Yu-Gi-Oh!, whose credits span movies, television and video games across a career of more than 30 years.

The court determined that using an actor’s voice without permission should be treated as an infringement of that person’s publicity rights. It is described as the first case of its kind in Japan, and it arrives as voice actors and other creative professionals confront a problem that is technically new but intuitively straightforward: if a recognizable performance helps draw an audience, who gets to authorize that use?

Tsuda alleged that narration on an anonymous TikTok account had been generated to imitate his voice. The account posted videos from July 2024 through the latter half of 2025. It was deleted before the ruling, so the court dismissed Tsuda’s request for an order requiring TikTok to remove the videos. That procedural outcome matters, but it does not erase the broader finding on voice rights.

What the court decided — and what it did not

The central legal takeaway is not simply that AI is controversial, or that every computer-generated voice resembles a particular actor. The finding is more specific: unauthorized use of an actor’s voice may violate publicity rights.

Publicity rights generally concern commercial value tied to a person’s identifiable persona. In this dispute, the argument was that an imitation did not merely supply generic narration. It traded on the appeal and recognizability associated with Tsuda as a prominent performer.

Tsuda maintained that the narration carried the deep, “lustrous” quality audiences associate with his delivery. His legal team argued the account’s popularity was connected to the novelty of sordid material being read in the voice of a well-known star. At its peak, the account reportedly had more than 200,000 followers.

TikTok disputed that characterization in court, maintaining the clips used a generic male voice and that any resemblance to Tsuda was open to debate. That disagreement illustrates a difficult practical point for future AI voice disputes: identity may be expressed through a bundle of traits rather than one easily measured audio feature. Pitch, cadence, articulation, resonance and performance choices can together make a voice feel recognizable, even where an imitation is not a literal recording.

The decision should therefore not be read as a blanket declaration that any low-pitched or polished synthetic narration belongs to a particular performer. Rather, it establishes that a human voice is not automatically free for exploitation simply because it is recreated by AI instead of copied from a sound file.

Why this matters to game and anime performers

For game audiences, voice performances are often inseparable from character identity. A familiar actor can give a rival, narrator, mentor or villain a distinctive presence across games, animation and related media. Tsuda’s long-running association with Seto Kaiba is a clear example of how a performer’s delivery can become part of a character’s cultural footprint.

That is why voice cloning presents a different question from using an anonymous stock-style narration voice. A synthetic performance can potentially leverage years of audience recognition without booking the performer, negotiating consent or compensating them. The issue is not restricted to official game production, either. Short-form clips, promotional material, parody-like content and other online formats can build views from the implication that a familiar voice is participating.

There is an important distinction between a performer’s recorded work and an AI-generated imitation. A recording may raise questions around copying a particular sound recording or performance. An imitation can avoid reproducing a specific clip while still trying to evoke the person behind it. This case addresses that latter concern through publicity rights: the commercial use of a recognizable human identity.

That distinction could be meaningful for creators and companies working around games. The absence of a copied line or lifted audio file does not necessarily settle the question of whether a voice has been appropriated. At the same time, the facts here centered on an actor whose voice and public profile were argued to be key draws for a sizeable social-media account. Future disputes will likely turn heavily on context, recognizability and use, rather than on a simplistic rule that all AI voices are prohibited.

A partial victory still sets a precedent

Tsuda did not receive the specific removal order he sought because the anonymous TikTok account was already gone. In immediate practical terms, there were no remaining videos for the platform to take down. But the ruling’s significance lies in the legal principle it recognizes, not in whether those particular posts remain online.

That makes the result a partial victory with wider implications. A deleted account can end one dispute’s direct remedy, but a court’s recognition that an unauthorized voice clone may infringe publicity rights can shape how later claims are argued. It may also encourage platforms, studios and advertisers to treat voice authorization as a distinct concern rather than assuming synthetic generation resolves it.

For performers, the decision gives added weight to a basic request: do not turn a recognizable working voice into a reusable digital asset without permission. For companies, it underlines the value of clear agreements that spell out whether recordings can be used for model training, digital replicas, synthesized dialogue, promotions or future projects not contemplated in the original job.

For audiences, it offers a reminder that “AI-generated” is a production method, not a substitute for disclosure or consent. A clip can be technically synthetic while still being designed to invoke a real person’s reputation.

Japan’s AI ambitions meet creator concerns

The ruling comes against a backdrop in which Japan has pursued a notably pro-AI national strategy. Yet the policy enthusiasm has existed alongside growing concern from anime voice actors and other creators about work, likenesses and creative output being absorbed by AI systems without adequate safeguards.

Those concerns extend beyond voices. Last year, the Japanese government asked OpenAI to ensure that anime and manga — described by the government as “irreplaceable treasures” — were not used to train the Sora 2 generator in a manner that breaches copyright. That request concerned training material and copyright, while Tsuda’s case focused on voice imitation and publicity rights. They are separate legal and technical issues, but both reflect the pressure to define what creators retain control over when generative systems can absorb patterns from existing cultural work.

Training is the process through which an AI system learns statistical patterns from material supplied to it. Inference is the later stage in which the trained system generates a result, such as a spoken narration. A dispute can focus on either stage, or on both. The Tokyo case is especially notable because it concerns the output: a synthetic voice allegedly made to sound like an identifiable actor.

That distinction is useful when discussing policy. Concerns over whether material was used to train a model do not necessarily answer whether a generated output misappropriates someone’s identity. Conversely, a system might be trained on lawfully obtained data while still being used in a way that misleadingly or commercially exploits a recognizable person. Each question needs its own analysis.

Practical implications for online creators and game-adjacent media

The facts of this case should not be inflated into a universal checklist, but they do point to sensible caution. Anyone commissioning synthetic narration that is meant to recall a named actor faces more risk than someone choosing a clearly non-identifiable voice. Popularity built around the perceived involvement of a celebrity performer may also make the use harder to characterize as incidental.

  • Seek permission before presenting a voice as, or deliberately near, a recognizable performer. This applies especially where the performer’s reputation is part of the attraction.
  • Do not rely on the “AI” label as a complete defense. The method used to make the audio does not automatically eliminate identity-based concerns.
  • Make contracts specific. Voice performers and commissioners benefit from addressing synthetic reuse, model training and digital replicas in clear terms.
  • Consider audience perception. If listeners are expected to believe, or be invited to believe, that a well-known actor is behind the voice, that context is likely significant.

The same caution is relevant across a wider entertainment sector in which performers move between animation, games, audio drama and screen work. As new creative ventures develop across comics, prose, audio and screen — such as this newly launched cross-media studio project — the boundaries between formats can become more porous, making voice and likeness permissions worth considering early.

Tsuda’s case does not resolve every question surrounding AI-generated performances in Japan. It does, however, put a clear marker down: a performer’s voice can carry protectable value, and recreating it without authorization is not necessarily consequence-free. In an industry where a single delivery can become inseparable from a beloved character, that recognition is likely to resonate far beyond one deleted TikTok account.