AI Regulation Tracker  /  Draft interpretive guidance

Japan's justice ministry drafted how existing law reads onto AI likeness and voice misuse

A draft final report from the Ministry of Justice works existing tort, publicity-right and unfair-competition doctrine through seven generative-AI fact patterns. It is a draft. It binds nobody. It is still the clearest signal yet of what a Japanese claimant would argue.

By Anthony Guerriero  ·  Published July 28, 2026  ·  Last updated July 28, 2026  ·  Standard news article

What did the Ministry of Justice actually publish?

A draft. That word is doing real work here, so it is worth being blunt about it before anything else. The file is captioned 取りまとめ報告書（案）, a draft final report, and the Japanese calendar date on its cover is 令和８年●月, with the month left as a placeholder. A document that does not yet know its own publication month has not been adopted.

What it contains is a legal organisation exercise. The study group was convened to work out how existing Japanese law answers questions that generative AI has made urgent, and the terms of reference say so directly.

生成ＡＩの普及等による肖像、声等の無断利用の事案が深刻化している等との指摘があることに鑑み、パブリシティ権等の権利の侵害に関する不法行為法の解釈適用等について、現行法及び判例法理を踏まえた法的整理の検討を行い、その結果を取りまとめる。

In light of observations that cases of unauthorised use of likeness, voice and the like arising from the spread of generative AI are becoming more serious, the Study Group shall examine the legal organisation of the interpretation and application of tort law concerning infringement of publicity rights and other rights, on the basis of existing law and case-law doctrine, and compile the results.

Ministry of Justice, statement of purpose for the study group

Read that clause slowly. 現行法及び判例法理, existing law and case-law doctrine. The mandate was never to write new law. It was to say what the law already is when the defendant is a diffusion model rather than a tabloid photographer.

Why does a draft that binds nobody still matter?

Because in Japan the answer to a publicity-rights question has historically lived in case law rather than in a statute, and case law is expensive to read at scale. There is no Japanese equivalent of a state right-of-publicity code you can hand to a client. Practitioners assemble the doctrine from judgments, from general tort principles, and from the Unfair Competition Prevention Act where a commercial identifier is involved.

A 140-page ministry document that walks seven concrete generative-AI fact patterns through that doctrine is a map of the arguments. It previews which theory a claimant reaches for. That is litigation posture, disclosed in advance, and counsel on the other side of a demand letter will have read it.

None of which upgrades it. A study group report is not binding authority, and a Japanese court is free to disagree with every page.

Who is affected?

Entertainment and IP litigators, talent agencies and their in-house counsel, voice actors, singers and actors, operators of voice- and image-synthesis products, unfair-competition practitioners, and rights-clearance managers. According to press reporting on the session, the scenarios include unauthorised use of actors' likenesses and of voice actors' voices in distributed audio, and use of a person's likeness in sexual imagery. The study group's published statement of purpose does not itself enumerate them.

How does this compare with the US, the EU and Denmark?

Voice and likeness cloning is one of the few AI harms where the United States moved first and moved by statute, which makes the comparison unusually informative. Japan is arriving at similar territory through interpretation of law it already had.

The structural point is the one worth carrying into a meeting. In the United States you ask which state's statute applies and whether the individual is alive. In Japan you ask which doctrine the claimant will plead, because the answer is not on a single page of a code.

What does this mean for a US brand cloning a voice or face for a Japanese campaign?

Treat the draft as a preview of the plaintiff's brief, not as a compliance checklist. If you are synthesising a voice or a face for a campaign that runs in Japan, the useful exercise is to read the scenarios and ask which one your campaign resembles, then assume a Japanese claimant will plead the theory the Ministry treats as strongest for that fact pattern.

Three practical consequences follow. First, consent documentation for synthetic replicas should be drafted to survive a tort analysis, not only a licence analysis, because the claim will likely sound in tort rather than in contract. Second, where the person is a commercially recognisable performer, the Unfair Competition Prevention Act angle is separate from the personality-rights angle and needs its own clearance thinking. Third, a US release signed under a state right-of-publicity statute is not automatically responsive to a Japanese claim, and counsel should not assume it travels.

There is also a timing question. A draft that reaches its fifth session is usually close to a final version, and the final version is what practitioners will cite. Work now that anticipates the final text is cheaper than remediation after it lands.

What the draft does not do

It does not create a cause of action. It does not set a penalty, a threshold or a registration duty. It does not establish voice as a protected interest as a matter of settled law. And it has not been adopted: the summary of proceedings for the 5th session was still marked as in preparation when this entry was written, so anyone relying on the report should confirm against the Ministry page rather than against press coverage.

Frequently asked questions

Does Japan's Ministry of Justice draft report create a new right over voice or likeness?

What did the Ministry of Justice publish on 27 July 2026?

Which scenarios does the draft report work through?

Has the guidance been approved?

Last verified: July 28, 2026