Japan adopts an emergency cyber package that has its AI Safety Institute rapidly test advanced AI models' attack abilities

Cabinet Secretariat of Japan · Policy report

Read the source (opens in a new tab)
What happened
Japan's Ministerial Meeting on Cybersecurity adopted a Cybersecurity Emergency Countermeasure Package on November 6, 2026, citing a run of unauthorized access and data leak cases.
Key points
  • The package says advances in AI mean gaps such as unpatched known flaws and misconfigured websites are now exploited in a short time and at large scale.
  • Japan's AI Safety Institute will work with AI developers to quickly assess advanced models' cyberattack abilities, and aims to share data on serious AI misuse with the UK AISI.
  • The government will deepen ties with large overseas digital firms and frontier AI companies to obtain attack information early and vulnerability details before disclosure.
  • Japan's NICT will start building STARDUST BD, a system to safely reproduce and study attacks by AI agents, including agents coordinating, evading controls and escaping to outside systems.
Evidence
The package as adopted, in Japanese; the wording here is a translation. It assigns planned measures to ministries, gives no budget figures, and does not name the frontier AI companies.

See it in the latest record