What happened
On 11 November 2026, a Japanese government working group on AI legal and institutional design met to consider making AI safety measures mandatory for high-performance AI business operators and introducing penalties for violations, to address risks such as cyberattacks and loss of control posed by high-capability AI. The government plans to outline the full institutional restructuring, including the content of the legal revision, by the end of 2026. Penalties would be directed at high-risk/frontier AI with significant risks, while general AI would continue to be addressed through existing laws and guidelines. Anthropic, OpenAI Japan and NTT participated, and the scope for regulation (including of major US IT firms) will be discussed.
Why it matters
Japan currently relies on a promotion-oriented AI law with no penalties; this working group signals movement toward binding, AI-specific obligations with enforcement teeth. If adopted, it would impose mandatory safety/incident obligations and penalties on frontier model providers serving the Japanese market, materially reshaping the APAC AI-governance landscape.
Action needed
Frontier AI model providers operating in Japan should monitor the government's year-end outline and prepare for potential mandatory safety measures, incident-reporting duties, correction orders and penalties.