Voice actors and AI voice clones: what the Tokyo and Shanghai rulings show, and what to keep

Flat illustration of an edited hire agreement beside the original, a risk report and a warning sign

Short answer: within one week in late September 2026, courts in Tokyo and Shanghai both found in favour of performers whose voices were copied by AI. Neither ruling is law in your country, but both show what helps a performer: a preserved copy of the offending audio, a comparison with the real voice, and records of where their voice was published. Keep dated originals of your recordings and contracts, and capture any suspected clone before it is deleted.

Last checked 12 November 2026. General information, not legal advice. Voice and likeness law differs by country and is changing. TimeBinder does not detect voice clones, and a dated record is not proof of authorship, ownership or that someone cloned your voice.

Table of Contents

Two rulings in one week

Tokyo, 30 September 2026. The Guardian reported that the Tokyo District Court ruled that the human voice has legal protection under publicity rights, in a case brought by actor Kenjiro Tsuda against TikTok over 188 videos on an anonymous account whose narration he said used a clone of his voice. The court said unauthorised use of a performer’s voice “can be considered an infringement on publicity rights”, but the protection applies where the purpose is exclusively to exploit its commercial appeal. It dismissed his request to remove the videos because the account owner had already deleted them after the suit was filed. TikTok argued the videos used a “generic male voice” and any similarity was subjective.

Shanghai, 29 September 2026. Trade publication Voice Over Herald, citing the court’s statement, reported that Shanghai’s No. 1 Intermediate People’s Court upheld 50,000 yuan in damages for a voice actress, surnamed Wang, whose voice a company had used in an AI-generated promotion. Wang had the audio notarised and commissioned a forensic lab comparison; the report says 24 of 28 measured acoustic indicators deviated by less than 10 percent. The company admitted the audio came from its own AI but said it could not confirm what the model had been trained on. The outlet, summarising an English-language legal analysis, says the court treated the company’s inability to show where its voice came from as the company’s problem. Press coverage elsewhere (Seoul Economic Daily) confirms the damages and the ruling that using a voice as training material without consent can infringe voice rights.

In the United States, a lawsuit against AI voice company Lovo, which alleges it used voice actors’ voices without permission, is still under way; a federal court allowed part of the publicity-rights claims to proceed, per Seoul Economic Daily. We make no comment on how it will end.

These are decisions on their own facts in Japan and China. They do not create rights in other places, and the Tokyo court refused the removal order. They are useful mainly as a picture of what evidence performers ended up relying on.

What evidence the cases turned on

  • A preserved copy of the suspected clone. In Shanghai the audio was notarised, in Tokyo the complaint described 188 specific videos. Accounts get deleted, as the Tokyo account was, so capture first.
  • A way to compare it with the real voice. Wang gave a sample recorded on the spot for a forensic acoustic comparison. Tsuda relied on the claim that the voice could only be his.
  • Evidence the performer’s voice was available to copy. The Voice Over Herald analysis says Wang had only to show a plausible route to her voice, since her work was publicly posted.
  • Evidence of use and commercial purpose. Revenue, subscribers and the promotional context.

What performers can keep

  1. Original session files and exports, with the dates they were recorded and delivered. Keep the raw WAVs, not only the mixes.
  2. Contracts, usage terms and consent emails, especially anything about AI, training or synthetic voices. If a client asks for a clause that allows AI use, ask what it covers and for how long, and get advice.
  3. A public-samples folder listing where your voice is published and when: demo reels, audiobooks, ads, with links.
  4. A capture log for suspected clones. For each: the link, the account, the date and time you found it, a downloaded copy of the audio, the page as displayed, and who you told. Do not edit the downloaded audio. A chain-of-custody log keeps this tidy, and screenshots have limits, covered in our screenshots guide.
  5. A dated fingerprint of each file. Make a SHA-256 of the originals with our hash checker (in your browser; the file stays on your device) and record a TimeBind of the set. See how to prove a file existed before a date.
  6. Talk to a lawyer before sending takedown demands or publicly accusing anyone. Notarisation is a Chinese practice; the equivalent preservation steps in your country may differ.

What a TimeBind does and does not show

A TimeBind is a dated, checkable record that an exact file existed no later than a recorded time and still matches. Your first Pooled TimeBind is free; see how it works. For a voice actor that can show that your original recordings, contracts and captured evidence are unchanged since the date you recorded them.

It does not show that a clip is a clone, who made it, that you own a voice, or that you consented or did not consent to a use. It does not stop anyone cloning a voice and cannot make a copy of your voice disappear. Whether an AI voice is similar enough to infringe is a matter for experts and courts, as the Shanghai forensic comparison shows. Our guide for lawyers covers how dated records fit into evidence handling.

Limits

Facts about the Shanghai ruling come from a trade-press report and a wire summary, not the judgment text. Voice-rights law differs sharply between countries: Japan has treated voice under publicity rights, China under personality rights, and the US relies on a patchwork of state laws. This is not legal advice.

Tools that can help

There is no affiliate link on this page: nothing here depends on a paid third-party tool. Our affiliate disclosure explains when we do use them.

Frequently Asked Questions

  1. Did a court really rule that AI voice cloning infringes rights?

    Two courts in Asia did, on their own facts. The Tokyo District Court (30 September 2026) said unauthorised use of a performer’s voice can infringe publicity rights where the purpose is to exploit its commercial appeal, and the Shanghai No. 1 Intermediate People’s Court (29 September 2026) upheld damages for a voice actress. Neither decision binds courts elsewhere.

  2. What should a voice actor keep if they suspect an AI clone?

    A downloaded copy of the audio, the link, account and date found, the page as displayed, your original recordings and contracts, and a note of who you told. Do not edit the files, and speak to a lawyer before acting.

  3. Does a blockchain timestamp prove my voice was cloned?

    No. A TimeBind records that an exact file existed no later than a recorded time and still matches. It can support your records of originals and captured evidence, but it does not show that a clip is a clone or who made it.

  4. Can I stop a company training AI on my voice?

    That depends on your contract and local law. Keep your contracts and usage terms, ask clients what AI clauses cover, and take legal advice. A timestamp does not prevent training.

Sources checked

Keep a dated record of your originals

Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. Click here to start, or see plans and pricing.