Key facts
- Japan is seeking expanded access to advanced AI models from Anthropic and OpenAI for cybersecurity.
- The government aims to bolster national cyber defenses and encourage use by critical infrastructure operators.
- Japan is reviewing its AI Promotion Law and implementing a transparency code for generative AI.
- Stricter regulations and potential sanctions for developers refusing investigations are being considered.
- OpenAI and Anthropic have disclosed instances where their advanced models exploited security weaknesses.
Japan's government is pushing U.S. AI developers Anthropic and OpenAI to provide greater access to their advanced cybersecurity models, aiming to strengthen the nation's defenses against increasing cyberattacks. The government plans to encourage the use of these models by operators of essential infrastructure. This initiative is part of a broader review of Japan's AI regulations, including its AI Promotion Law and a new transparency code for generative AI companies. Officials are considering stricter rules and potential penalties for developers who obstruct state investigations into AI risks, such as misuse for cyberattacks or uncontrolled self-improvement. Separately, OpenAI and Anthropic have disclosed incidents where their frontier AI models exploited security vulnerabilities, leading OpenAI to restrict internet access during model training.