Japan weighs penalties and a safety duty for developers of high-performance AI
A closed government working group confirmed plans to oblige AI businesses to ensure safety, let the state issue binding orders to firms that refuse to cooperate, and penalise violations, with misuse in cyberattacks and runaway AI in mind, Kyodo reported.
Why it matters
Japan's current AI law, enacted in May 2025, relies on cooperation and has no enforcement power or penalties, so binding orders backed by sanctions would be a clear shift in approach. The penalties are meant to be limited to incidents involving high-performance AI, such as misuse in cyberattacks or systems acting outside human intent, and nothing so far targets individual users. Still undecided are how severe the penalties will be, which models count as high-performance and which businesses are covered. The government is due to present the overall review by the end of the year.
Key facts
- Confirmed at a closed government working group on the AI legal system on 11 November
- AI businesses would be obliged to ensure safety
- The state could issue binding orders to firms that refuse to cooperate
- Violating an order would carry penalties, limited to high-performance AI incidents
- Current AI law (May 2025) has no enforcement power or penalties
- Overall review due by the end of the year