Voice Is a Right: Tokyo's AI Ruling and What Boards Must Audit

On 30 September 2026 Tokyo recognises the voice

On 30 September 2026 the Tokyo District Court recognised that a person's voice falls within protected publicity rights in Japan. Judge Aya Takahashi wrote that the human voice is a symbol of individual personality, in the same way a likeness is. This marks the first time a Japanese judge has reached that conclusion.

The same decision dismissed the claim brought by voice actor Kenjiro Tsuda, who had asked TikTok to remove videos carrying a voice clone generated by artificial intelligence.

The suit was filed in November 2025 and stands as Japan's first case on a voice rights violation produced with generative AI. Tsuda voices lead roles in franchises such as «Jujutsu Kaisen». The ruling applies to the Japanese jurisdiction and takes effect from 30 September.

The delta: from name and likeness to timbre

Until this decision, publicity rights in Japan covered the name and physical appearance of well-known individuals, with the aim of governing the commercial exploitation of identity. The field took shape mainly in the courtroom, ruling after ruling, rather than through dedicated legislation.

The voice sat outside that list, left to indirect instruments: defamation, unfair competition, copyright in the recording.

The Tokyo court moved the boundary. A person's timbre enters the perimeter of protected commercial identity, and its use without consent becomes a matter of tort. The Kyodo dispatch carried by Nikkei Asia^[1] places the ruling within the global line of voice cloning cases.

For an organisation generating synthetic voices, the inventory of rights to clear gains one line: the timbre of the person being imitated.

Right recognised, remedy still open

The takedown claim fell on procedural grounds: the videos had already been deleted. The bench recognised the right and declared the interest in the measure exhausted.

The outcome leaves organisations with a clear rule and an empty precedent on remedy. A platform receiving a cease and desist letter today knows that the voice is protected; the breadth of the takedown order awaits another judgment. MLex, a specialist legal and regulatory intelligence service, followed the case in its intellectual property channel^[2].

Two readings circulate among counsel. The first treats the recognition as a passage in the reasoning, with limited force until another bench applies it to an actual order. The second treats it as a principle usable right away in cease and desist letters and contractual clauses.

The question on the right has found its answer. A second question has opened.

The numbers the board needs to read

The account named in the proceedings published 188 videos narrated by a voice imitating the actor's. The posts span July 2024 to September 2025. The profile held roughly 210,000 subscribers, with an average of 1.47 million views per video, according to the account given by the Japan Times^[3].

Estimated revenue for whoever ran the account ranges between 500,000 and 750,000 yen a month, around 3,100 dollars at the lower end. Many subscribers commented that the narration sounded like Tsuda.

These figures matter in the boardroom for an accounting reason: they measure the economic value captured from a third party's vocal identity. A risk sheet on cloning starts here, from the estimate of transferred earnings and the duration of the exposure.

Fourteen months of publishing and over a million average views per video describe an industrial operation, far from the isolated episode.

The platform's defence and the recognisability threshold

TikTok argued in court that the video maker had trained the model on a friend's voice and that the output remains a generic male voice. The defence touches the technical heart of the matter: recognisability.

The recognisability threshold decides the reach of the rule. A synthetic voice evoking a famous timbre stays lawful up to the point where the audience attributes that sound to a specific person. The subscriber comments naming the actor become evidence of attribution.

This dynamic shifts the internal burden. Whoever produces synthetic audio gathers evidence of the provenance of the vocal data before release, with signed consent from the person who lent the voice. The «generic timbre» defence holds as far as the upstream documentation holds.

The operational lesson is short: the harm arises from audience perception, ahead of the model's code.

The governance signal

The governance signal: the voice moves from the ground of taste to the ground of rights to clear, and due diligence on vocal data becomes a contractual clause. Whoever buys synthetic voices also buys the upstream chain of consent.

Outside Japan, part of the architecture already exists. Article 50 of European Regulation 2024/1689 imposes a transparency obligation on synthetic content imitating real people and is applicable from 2 August 2026. In the United States, the Tennessee ELVIS Act protects name, voice and likeness from 1 July 2024, with a state-level and fragmentary approach.

Fragmentation is the model, and the Tokyo ruling extends it to Asia. A multinational group today governs three separate regimes over the same audio track: a label in Europe, a state-level tort in the United States, a publicity right in Japan.

The cost of compliance drops for whoever maps the voices already in production now, calmly.

Three decisions for the board

The Japanese decision moves voice cloning from the ethical plane to the accounting plane. Three points need closing in writing, with a named role and a review date.

1. General Counsel and Chief Compliance Officer: which chain of consent covers every synthetic voice in production, by contract, by date and by jurisdiction?
2. Chief Risk Officer: does the risk register include third-party voice cloning among identity torts, with exposure estimated in currency?
3. Board Audit & Risk Committee: which role, named in writing before release, answers for the use of generated voices, and in which minutes does it appear?

The audit remains mandatory; the perimeter has changed. A control calibrated on copyright in the recording now sits out of scale against the timbre of a living person.

The CEO closes one point: which product lines use synthetic voices of real people, and under which licence. A framework that avoids naming the accountable lead produces documentation; governance arrives from the name written down before release.

Regulatory horizon

In Japan the ruling takes effect from 30 September 2026 and stands as the first judicial recognition of the voice among publicity rights. Appeal rests with the parties. The perimeter of the remedy awaits a case with the content still online.

In the European Union, Article 50 of Regulation 2024/1689 is applicable from 2 August 2026. The obligations on high-risk systems carry a later date, moved to December 2027 by the simplification package still under discussion among the institutions.

In the United States, voice protection runs through state law, with Tennessee in force since 1 July 2024 and other states in the queue. A coherent federal rule remains distant, and the realistic calendar looks to the end of the decade.

Whoever maps consent, licences and named roles now banks the advantage that enforcement will make visible in eighteen months.