The Japan Actors Union dropped its first formal rules on generative AI voice use this week, and the message is simple: ask at every step or get blocked. The 22-article guidelines, released October 15 after a Tokyo press conference, require written consent before any voice data hits training, before a model generates new audio, and before that output gets published or sold. No blanket buyouts. Consent maxes out at one year in most cases.

AI-generated performances count as a fresh "real performance," so actors get separate royalties on top of the original session fee. Automatic translations into other languages need fresh permission per language plus extra pay for global reach. The union also drew a hard line on the dead: recreating deceased seiyuu voices is banned outright except for rare continuation cases where the performer had already approved the project while alive, with family, agency, and royalty checks required. Dragon Ball legend Masako Nozawa, 89 and vice-chair of the union, called unauthorized clones "the same as theft" and said AI copies "have no heart." Bin Shimada, Broly's voice, described hearing an AI version of his character singing Ado's "Usseewa" and said it left him stunned then furious.

The union represents roughly 2,600 actors and voice actors. Japan still lacks a dedicated "voice right" statute, so these rules are voluntary industry standards meant to shame companies into compliance while the group pushes the government for actual law. Over half of Japanese developers already use generative AI in some capacity, mostly for art and text, which makes these guardrails land right as the tech scales into voice work for games and anime.