JapanJapan asks Anthropic and OpenAI to widen access to cybersecurity AI models ahead of next week's cyber package

Japan's government has asked Anthropic and OpenAI to expand access to their advanced security AI models, Global Economic reported on Oct. 31, citing Nikkei Asia.

Tokyo wants to use the models to check vulnerabilities in critical infrastructure such as power, gas and telecoms, and in private systems. It gained official access to Anthropic's Claude Mythos in June and has early access to OpenAI's GPT-5.5-Cyber, which it is comparing for defensive performance. A national cybersecurity package due next week includes deeper cooperation with US AI firms.

Cyber Security Cloud counted about 450 million attack detections or unauthorized access attempts against Japanese websites in September, up 36% from August. Cybersecurity Minister Toshiharu Furukawa met Anthropic co-founder Jack Clark on Oct. 13, and Prime Minister Sanae Takaichi told the cabinet on Oct. 27 to work with foreign authorities and global tech firms to identify attackers and fix vulnerabilities fast. A senior official said basic security such as software updates still matters for private firms.