The guidelines cover four points. First, permission must be granted at "every stage" of the generative AI process, meaning AI companies would need actors' consent before training models on their performances. Second, AI replications of an actor's work count as a new performance, so royalties or compensation would be owed to the actors involved.

Third, AI-powered translations that automatically convert a performance into another language must maintain the "artistic integrity" of the original and may not be used without permission in overseas regions. Finally, cloning a deceased actor's voice is banned unless the actor gave express permission while alive, and even then an AI company would need approval from the actor's family and affiliated agency or company, plus royalty payments.

The union says it will push for these guidelines to become industry-wide rules and will encourage the Japanese government to legislate protections for "voice rights". It stressed that "technological advancements will not stop, but the 'living, breathing human expression' and a 'respect for rights' must always remain at its core."

Important caveat: these are guidelines, not laws, so they are not binding on AI companies. They nonetheless add to mounting pressure on the generative AI industry in Japan. There is precedent for the underlying argument: voice actor Kenjiro Tsuda, known for Jujutsu Kaisen and Chainsaw Man, sued TikTok to remove AI voice imitations of him. He lost the case, but it established that a person's voice is subject to protection under publicity rights laws. Western performers' collectives have raised similar concerns, with one describing the technology's impact as "catastrophic".