Tokyo reportedly wants more of the restricted cyber-focused models it already uses for vulnerability checks across government and finance
Japan wants a bigger seat at the table where the world’s most tightly guarded AI models are handed out.
The country is asking Anthropic and OpenAI for more access to their advanced models. Nikkei reported on October 30, 2026 that Tokyo plans to formally request further use of these high-performance systems, specifically to run vulnerability inspections.
What Japan already has, and what it wants next
Japan is not starting from zero. Earlier agreements in 2026 already gave Japanese government bodies and financial institutions some access to two specialized models: Anthropic’s Claude Mythos and OpenAI’s GPT-5.5-Cyber.
The Anthropic piece arrived in early June 2026. The Japanese government and a select group of financial institutions gained access to Claude Mythos Preview as part of an expansion of Project Glasswing.
That expansion stretched the program to approximately 150 organizations across more than 15 countries.
The OpenAI track ran through the finance ministry. OpenAI granted Japanese banks access to GPT-5.5-Cyber after Finance Minister Satsuki Katayama held discussions with OpenAI executives in May 2026.
The US government helped coordinate that arrangement.
Now Japan wants more. According to the Nikkei report, the next request would focus on expanded provision of these models for checking systems for weaknesses. A related national cyber package is expected to be announced soon, aimed at strengthening international cooperation on cybersecurity.
The digital minister’s charm offensive
Digital Minister Toshiharu Furukawa has been doing a lot of the legwork. By mid-October 2026, he had met with Anthropic’s Jack Clark to push for deeper cooperation on identifying vulnerabilities.
He also pitched Japan as a location for data centers.
Why these models are on a short leash
The controls on these systems come down to one uncomfortable capability. They can spot software vulnerabilities that human experts might miss.
That dual-use problem shapes everything about how access gets granted. The ability to uncover flaws that evade human detection raises security concerns on both the offensive and defensive side, and it complicates Japan’s effort to get more of it.
What this means
For Japan’s financial sector, the stakes are direct. Banks were among the first Japanese institutions to receive access to both GPT-5.5-Cyber and Claude Mythos Preview, which suggests policymakers see finance as a priority target worth defending with the strongest available tools.
For the AI labs, Japan’s push is a test of how they ration their most sensitive products. Project Glasswing’s reach of approximately 150 organizations shows Anthropic is willing to widen the circle, but carefully.
The US government’s role in coordinating the OpenAI arrangement with Japanese banks is also worth watching. If Washington continues to act as a broker for allied access, these models start to look like a component of alliance-level security cooperation, not just a commercial product line.
The data-center proposal adds another layer. Hosting infrastructure for frontier AI firms would give Japan a stronger relationship with the companies controlling these tools. Whether Anthropic takes Furukawa up on the offer has not been confirmed.
The upcoming cyber package is the next concrete marker. Its emphasis on international cooperation suggests Tokyo sees cybersecurity as something it cannot handle alone, and that access to foreign frontier models is now part of the national defense toolkit.
