Four courts, four different answers on AI voice cloning consent in 2026

What happened, in order

Five dated events, across four countries, each run through a different existing legal theory because none of these countries has a statute written specifically for AI voice cloning.

Early September 2025, Berlin. A German court ordered a YouTuber to pay voice actor Manfred Lehmann, Germany's dubbing voice for Bruce Willis since the 1980s, 4,000 euros in damages for using an AI clone of his voice in political commentary and advertising videos without consent. The ruling rested on Lehmann's personality rights, a long-established German legal concept, not on any AI-specific statute.

30 January 2026, Paris. Eight French dubbing actors, including Richard Darbois (the French voice of Harrison Ford and Richard Gere) and Brigitte Lecordier (the French voice of Son Goku and Buzz Lightyear), sent formal demand letters to two US-based platforms, VoiceDub and Fish Audio, which sold subscription access to AI clones of their voices. The letters, drafted by lawyer Jonathan Elkaim and reported by Mac4ever, invoked a right-to-privacy theory, demanded removal within eight days, and sought 20,000 euros in damages per actor.

By 6 April 2026. Journal du Geek reported that VoiceDub had removed the flagged clones quickly and Fish Audio had followed after taking longer, and that dozens of French dubbing actors beyond the original eight had separately obtained removal of clones using their voices. No lawsuit was filed. The outcome was reached entirely through demand letters.

2 August 2026, EU-wide. The AI Act's Article 50 transparency obligations began to apply, requiring providers of systems generating synthetic audio to mark the output as AI-generated. This is a labeling duty, not a consent requirement: it says nothing about whether cloning a voice in the first place needs the speaker's permission.

30 September 2026, Tokyo. The Tokyo District Court ruled on a suit by voice actor Kenjiro Tsuda, known for voicing Kento Nanami in Jujutsu Kaisen, against the anonymous operator of a TikTok account that had posted more than 180 videos narrated by an AI clone of his voice, reaching over 200,000 followers by November 2025. Presiding Judge Aya Takahashi found the use infringed Tsuda's publicity rights, reasoning, as reported by the Associated Press, that "the human voice is symbolic of individual personality, just like one's portrait." The court dismissed the request to delete the videos, because the account had already closed on its own.

What each legal theory actually delivers

The same underlying harm, an AI clone of someone's voice used without permission, produced four different remedies because four different bodies of law were used to reach it. None of the four is a purpose-built voice-cloning statute.

Only one of the five produced a remedy with teeth, and it was the oldest doctrine, not a new one. Germany's personality-rights case is civil law that predates AI by decades, applied to a new fact pattern, and it is the only row in Table 1 that ends in money changing hands and a legally binding order. The newest attempt, Japan's publicity-rights recognition, produced a legal first ("the human voice is symbolic of individual personality, just like one's portrait") but zero enforceable consequence, because the harm had already stopped by the time the court ruled.

France shows a demand letter can do what a lawsuit threatens to. Two platforms removed the actors' voices within the letters' window without a judge ever hearing the privacy-rights argument. That is a good outcome for the eight actors, and a bad data point for anyone trying to learn what French law actually requires, because nothing was adjudicated. The theory remains untested.

The EU's rule does not touch the actual question. Article 50 requires the output to be marked as synthetic. It says nothing about whether the speaker had to agree to the clone being made. A fully labeled, fully disclosed AI clone of someone's voice, made without their consent, is compliant with Article 50 and potentially actionable under none of the frameworks in this table outside Germany, Japan, or wherever a plaintiff can reach a sympathetic personality- or publicity-rights doctrine.

Why the same harm gets four different outcomes

None of these four legal systems has a rule that asks the question a dubbing buyer actually needs answered: did the speaker consent to this specific clone, for this specific use? Each one answers an adjacent question instead, and the adjacent question determines what remedy is even available.

This matters for anyone buying or building AI dubbing. A clone that would trigger a damages award in Germany, be removed voluntarily by a platform in France, be recognized but left unremedied in Japan, and face no federal claim at all in the United States is, mechanically, the same clone. The outcome depends on where the speaker can sue, not on what was done to their voice.

What this means if you buy or build AI dubbing

Written consent is the only thing that works in all four jurisdictions. Every case in Table 1 exists because the speaker did not agree to the clone. A signed, scoped consent record, naming the use, the duration and the platform, is the one artifact that defangs personality-rights, privacy-rights, publicity-rights and (once it exists) NO FAKES Act claims simultaneously. None of the four legal theories has an exception for "we labeled it," per the EU's Article 50 column above.

A platform's voluntary takedown is not legal clearance. VoiceDub and Fish Audio removing the French actors' clones resolved the specific complaint, but it tested nothing. A platform that complies once under threat has not established it had the right to clone the voice in the first place, and a different actor without a lawyer sending letters may get a different response.

Do not treat the US as a safe harbor because the NO FAKES Act hasn't passed. The absence of a federal right of publicity does not mean voice cloning is unregulated there; it means protection depends on whichever state's law applies, inconsistently. Track S.4591 and H.R.8915 on Congress.gov rather than assuming the current gap is permanent: the bill has bipartisan sponsors and industry support from SAG-AFTRA, Google, OpenAI and Amazon, and committee passage was unanimous.

Labeling and consent are two separate compliance tracks. Meeting Article 50 in the EU proves the output is marked as synthetic. It proves nothing about whether the speaker agreed to be cloned. Treat them as two checklist items, not one, because a fully labeled unauthorized clone is still exposed everywhere outside the EU's transparency rule.

Check it yourself

Each ruling and filing below is a primary or wire-service source with a date. The bill text and committee status are checkable directly on Congress.gov.

# 1. Current status of the federal bill, straight from Congress's own system
#    (replace the session/congress number as needed; this is the 119th Congress)
curl -s "https://www.congress.gov/bill/119th-congress/senate-bill/4591" -A "Mozilla/5.0" \
  | grep -io "latest action[^<]*" | head -5
curl -s "https://www.congress.gov/bill/119th-congress/house-bill/8915" -A "Mozilla/5.0" \
  | grep -io "latest action[^<]*" | head -5

# 2. Read the EU AI Act Article 50 text directly, it is short
#    https://artificialintelligenceact.eu/article/50/

# 3. The Tokyo ruling and the Berlin ruling are both covered by wire services,
#    not just blogs. Cross check the AP account against a second outlet dated
#    the same week before citing the quote from Judge Aya Takahashi:
#    https://ny1.com/nyc/all-boroughs/ap-top-news/2026/10/01/tokyo-court-grants-legal-protection-to-human-voices-in-ai-clone-case
#    https://www.thestandard.com.hk/world/article/344249/Japan-court-rules-human-voice-is-protected-in-TikTok-AI-case

# 4. The French case has no court filing to check, only the demand letters
#    and the platforms' response, reported by two independent French outlets:
#    https://www.mac4ever.com/ia/194464-huit-doubleurs-francais-declarent-la-guerre-aux-cloneurs-de-voix-ia
#    https://www.journaldugeek.com/2026/04/06/doublage-et-ia-les-voix-francaises-refusent-detre-clonees-sans-permission/

If you are drafting a consent record for a dubbing vendor, the one line worth adding after reading these five sources: name the jurisdiction the speaker can sue in, not just the jurisdiction the buyer operates in. That is the variable that actually decided every case above.

What would prove this wrong

The claim under test is that, as of October 2026, no jurisdiction examined here has a legal rule that directly tests consent to a voice clone, only adjacent doctrines repurposed for the fact pattern. This is wrong if, by 1 July 2027, the NO FAKES Act (or a successor bill) is signed into US federal law, since its text explicitly creates a licensable property right conditioned on the voice owner's authorization, which is a direct consent test rather than an adjacent one. It is also wrong, independently, if a French, German or Japanese court issues a published ruling that turns centrally on whether consent was given, rather than on personality, privacy or publicity exploitation as a proxy for it.

A second, narrower prediction, marked as judgment: the Tokyo ruling's zero-remedy outcome will be cited by future Japanese plaintiffs' lawyers as a reason to seek an injunction before an account can close voluntarily, rather than relying on after-the-fact publicity-rights litigation. If no published commentary makes this argument by the same date, that reading was wrong.

FAQ

Is AI voice cloning for dubbing legal and consented in 2026?

It depends entirely on jurisdiction, and none of the four most-discussed cases of 2026 relied on a law written for voice cloning. Germany used personality rights to award a dubbing actor 4,000 euros. France never reached a court: two platforms deleted cloned voices after demand letters that threatened a privacy-rights suit. Japan's Tokyo District Court recognized a publicity-rights infringement on 30 September 2026 but ordered no remedy, because the infringing account had already closed. The United States has no federal right of publicity at all; the NO FAKES Act, which would create one, passed a Senate committee in June 2026 and has been stalled since early October.

What did the Tokyo court actually rule in the Kenjiro Tsuda case?

On 30 September 2026 the Tokyo District Court held that an unauthorized AI clone of voice actor Kenjiro Tsuda's voice, used to narrate over 180 TikTok videos under the account name Nanami, infringed his publicity rights. Presiding Judge Aya Takahashi said the human voice is symbolic of individual personality, just like one's portrait. The court dismissed Tsuda's request to have the videos deleted because the account had already been closed, so there was nothing left to order removed.

Did the French dubbing actors win a court case against VoiceDub and Fish Audio?

No court ruled. Eight French dubbing actors, including Richard Darbois and Brigitte Lecordier, sent formal demand letters dated 30 January 2026 to VoiceDub and Fish Audio, invoking a privacy-rights theory and threatening suit within eight days unless the platforms removed the cloned voices and paid 20,000 euros each. VoiceDub removed the flagged content quickly; Fish Audio took longer but complied. By April 2026, press coverage described dozens of dubbing actors having obtained removals through the same letter-based route, with no lawsuit filed.

Is there a federal law against cloning someone's voice without consent in the United States?

Not yet. The United States has no federal right of publicity; protection depends on a patchwork of state laws. The NO FAKES Act (S.4591 and H.R.8915), which would create a federal, licensable property right in a person's voice and visual likeness, was reintroduced on 20 May 2026 and passed the Senate Judiciary Committee unanimously on 18 June 2026. As of 7 October 2026 it was stalled in the Senate until at least November after Senator Ted Cruz objected to a unanimous consent request over free-speech concerns.

Sources

1. OECD.AI AI Incidents Monitor, German court ruling against AI voice cloning of Bruce Willis' dub actor, incident dated 3 September 2025. Manfred Lehmann, Berlin court, €4,000 damages, personality rights.
2. Mac4ever, Huit doubleurs français déclarent la guerre aux cloneurs de voix IA, 30 January 2026 demand letters, Richard Darbois, Brigitte Lecordier, VoiceDub, Fish Audio, lawyer Jonathan Elkaim.
3. Journal du Geek, Doublage et IA: les voix françaises refusent d'être clonées sans permission, 6 April 2026. VoiceDub and Fish Audio removals, dozens of additional actors.
4. European Union, AI Act Article 50, transparency obligations, applicable from 2 August 2026.
5. Associated Press via NY1, Tokyo court grants legal protection to human voices in AI clone case, published 1 October 2026, ruling dated 30 September 2026. Kenjiro Tsuda, Judge Aya Takahashi, TikTok account "Nanami."
6. The Standard (Hong Kong), Japan court rules human voice is protected in TikTok AI case, 30 September 2026.
7. Manatt, Congress Reintroduces the NO FAKES Act: What's New in the 2026 Bill; Senator Marsha Blackburn's office, press release on Senate Judiciary Committee passage, 18 June 2026; Daily Journal, NO FAKES Act dead for now as Cruz objects to unanimous consent request, 7 October 2026 (headline and summary only, article paywalled).

Related BLOMEGA guides: Consented AI training data providers · A dubbed minute costs $0.33 to $9.00, and the watermark discount is gone.