Japan targets late November for Defense AI Vision, with humans accountable

Japan’s Defense Ministry aims to publish a new policy called the Defense AI Vision around the end of November. Defense Minister Shinjiro Koizumi announced the target at a Nov. 15 press conference in Kurihama, Yokosuka. The document will set out how the ministry and the Self-Defense Forces plan to use AI and update the basic policy they issued in July 2024.

The target gives research partners and suppliers of data and cloud services an approximate date to review the policy. Koizumi outlined three priorities: keep judgment and responsibility with people, make practical use of data, and introduce AI quickly while protecting safety and information. The November target covers publication of the policy. The briefing gave no dates for individual projects or changes to contract terms.

Where people retain responsibility

The 2024 policy already treated AI as a tool for human judgment. It called for staff to identify the problem first and then assess whether AI could help. Making administrative work more efficient was one example of where the ministry wanted to apply that approach. The policy also called for sharing data and expertise across organizational lines and working with outside businesses and research institutions.

The ministry followed the policy with a June 2025 guideline for responsible AI use in its research and development. Covered projects are grouped by risk. Projects may undergo legal, policy and technical reviews or an assessment by the team in charge, depending on their classification. Review criteria include whether documents spell out what people are responsible for, what users need to know about a system’s limits, and whether the people examining a project can later verify the data and methods used to build it.

Koizumi said the new vision would focus on bringing AI and data into routine use across the ministry and the Self-Defense Forces. The 2025 guidance identifies points in a project when reviews can occur, such as before design approval or testing. The scope of the new policy will matter when staff and partners need to determine who checks an AI result or makes a decision once a service is in use.

How the ministry would protect data in the cloud

Asked about cloud infrastructure, Koizumi said safeguards should reflect the sensitivity of the information it handles. He stressed Japan’s ability to operate and manage the environment independently while giving the ministry and the Self-Defense Forces lasting, secure access to data. More sensitive information would require preparation and standards before it could be handled in the cloud, he said. He gave no timetable for introducing those standards.

The groundwork for sharing data goes back to the 2024 policy. It called for identifying the records that units needed to share and standardizing data formats. The ministry also discussed bringing in people with AI and data experience from outside the organization. Koizumi’s remarks put the emphasis on protecting records when staff use them and on keeping the systems available over time.

At an Oct. 20 press conference, Koizumi told reporters the ministry would consider work on Japan’s three national security documents as it reviewed the AI policy. On Nov. 15, he also discussed training staff alongside providing a secure environment for continued AI use.

To explain why the ministry was revisiting its policy after about two years, Koizumi pointed to changes in Japan’s government-wide AI plan. The Cabinet adopted its first basic plan on Dec. 23, 2025, and a second version on July 14, 2026. The Defense AI Vision addresses how the ministry and the Self-Defense Forces use AI in their own work.

For suppliers and research partners, the public documents currently available are the July 2024 policy and, for research covered by it, the June 2025 guideline. The outstanding questions for those partners are which activities the vision will cover and how much of the planned cloud policy the ministry will spell out in the document.