Japan Actors Union publishes voice-AI rules on consent, fees, and dubs

On 15 October 2026 the Japan Actors Union, 日本俳優連合 (Nippairen), published a guideline on generative AI and performers’ voices. The announcement is timestamped 2026-10-15T04:25:32Z and says about 2,600 actors and voice actors belong to the cooperative. NHK WORLD-JAPAN also says about 2,600 members. The linked PDF, “音声に関する生成AI利用ガイドライン,” version 1.0, is dated 15 October 2026 on the cover. Quotations from the Japanese are our translations.

The text is a guideline the union asks companies to follow. It is not a statute. The announcement says Japan has no explicit law that comprehensively protects a person’s “voice” itself, a so-called “voice right.” That sentence is the union’s. The Asahi Shimbun, at 16:50 JST on 16 October, reports that the industry ministry has begun considering changes to the Unfair Competition Prevention Law. No ministry paper was opened. The guideline’s preamble says Copyright Act Article 91 assumes a recording of the performance itself and does not necessarily cover learning a voice in order to generate a new one. That is the union’s reading.

Consent, the one-year term, and the fee

Article 8 forbids a blanket or broad prior consent, including a buyout, and treats that kind of clause as invalid in a contract, a rider, or boilerplate. Oral consent, implied consent, and a yes after the fact are treated as invalid too. Another project, series, season, or unplanned medium needs its own written consent when the use arises.

The announcement, citing Articles 7, 8, and 9, says permission is required at learning, at generation, and at publication. Article 8 requires a separate written consent at each of those stages, and a further consent for each ad campaign. Article 7 requires advance notice of the project, the technique, the content, whether voices are mixed, and the withdrawal procedure.

Article 12 caps consent, in principle, at one year. Continuing means a new written consent after the fee is discussed again. A longer term is allowed only for a long production, such as a game or a series, and only for the span that work needs. An open-ended consent is treated as invalid. Article 11 says a refusal or a withdrawal must not affect later casting or auditions.

Article 16 requires an AI fee, agreed before use, separate from the ordinary appearance fee. If there is no agreement, the operator is not to proceed. The PDF lists a learning fee, a use fee, an advertising fee, and a multilingual fee. It names no yen amount.

Article 17 says that when a performer asks to withdraw and the operator agrees, the voice data and prompts are deleted within 30 days, and later use of a model trained or fine-tuned on that data stops. Deletion follows an agreed withdrawal.

Dubs, full replacement, and the dead

Article 14 says animation, foreign-film and drama dubs, and other Japanese-language acting should in principle not be done by replacing the performer with AI. An exception needs permission from the performer, the voice rights holder, the work’s rights holder, and the audio producer, in an individual written contract that includes the fee. Small technical edits inside an agreed take, such as noise removal or a mouth-sync tweak, are allowed. Article 15 requires a separate consent for each language.

Article 20, paragraph 3, says a fully synthetic performer used in place of a human one needs prior consultation and written approval from a body the PDF provisionally names 音声AI取引適正化委員会. Asahi renders that as a “Voice AI transaction fairness committee” and says the details and the timing come later.

Article 6 says operators in principle must not build, keep, or use a dead performer’s AI voice. The exception is a sequel already in concrete production at death, or a project the person had agreed to while alive, and only the minimum a living performer cannot cover. It still needs prior written consent from the statutory heirs and the former agency, for a stated purpose, scope, and period, and a fee at least equal to the living rate. New works, ads, and events are excluded. If neither family nor agency can be found, the exception stays closed.

What a performer can ask for

Article 20 says a breach lets the performer demand an immediate stop to distribution and sale, and irreversible disposal of the trained model, including the weights. The operator is then to pay special damages equal to the sales of that business, back to the day the unauthorized use began. NHK’s 16 October story reports the same demand. These are requests in a guideline. They are not a court order.

Asahi reports a video message at the 15 October Tokyo news conference from Masako Nozawa, the Japanese voice of Son Goku in Dragon Ball: “I hope everyone follows the rules so that both people and AI can enjoy a future where they can happily coexist.” NHK also reports a video message to the union and calls her vice chair. The video file was not opened. Asahi says two industry groups “reportedly support” the standards. Their own sites were not the source for that line.

Meloni’s EUIPO application is a trademark filing, a different instrument. The Nippairen text asks for a fresh written yes, a separate fee, and an end date.