Japan MOD AI utilization policy and ATLA responsible AI R&D guideline

Japan’s live July 2024 title is 防衛省ＡＩ活用推進基本方針. ATLA’s later English edition of the R&D guideline refers to that document as the MOD Artificial Intelligence Strategy (Ministry of Defense, July 2024). Official pages still say 人工知能 / ＡＩ / Artificial Intelligence. This article uses those names. Super Intelligence (SI) is this site’s own-voice term for current U.S. executive-branch policy; it does not rewrite a Japanese instrument. ATLA’s English PDF is marked Provisional translation for reference purposes only.

What the July 2024 policy published

The 2 July 2024 press conference says the Ministry formulated for the first time the 防衛省ＡＩ活用推進基本方針 as a compass for promoting AI use.

The summary and full PDF name seven priority fields: ①目標の探知・識別、②情報の収集・分析、③指揮統制、④後方支援業務、⑤無人アセット、⑥サイバーセキュリティ、⑦事務処理作業の効率化. The same pages say use is not limited to those seven fields and that trying first also matters where AI can help solve a defined problem.

On human involvement, the policy says ＡＩが行うのは人間の判断のサポートであって、その活用に当たっては人間の関与を確保する必要がある. For risk, it takes as reference the ＡＩ事業者ガイドライン ideas of ①人間中心、②安全性、③公平性、④プライバシー保護、⑤セキュリティ確保、⑥透明性、⑦アカウンタビリティ. It also says the Ministry will formulate its own guidelines for research and development of AI-using equipment.

The policy is a Ministry compass, not DoD Responsible AI and not NATO PRUAI. See Australia’s Policy Settings and Canada DND/CAF for other national Allied documents that still use the published name AI.

What ATLA’s R&D guideline adds

ATLA’s landing titles the later document 装備品等の研究開発における責任あるＡＩ適用ガイドライン. It is the Ministry of Defense and Self-Defense Forces’ own guideline showing the concept of responsible AI application in equipment R&D, issued after the July 2024 防衛省ＡＩ活用推進基本方針. The same page publishes an English edition: Guideline for Responsible AI Application in Research and Development of AI-Equipped Defense Systems (Ver. 1). A 6 June 2025 notice announces that formulation.

The English edition says the Guideline aims to materialize a series of commitments of the Ministry of Defense and Self-Defense Forces and provide a framework for appropriately managing and mitigating AI risks in the context of research and development. That is intended to ensure predictability for businesses participating in research and development. Scope is research and development projects from the conceptual stage to the research and development stage. The definition of AI in the Guideline adheres to the Strategy and refers to machine learning software and programs.

Requirement A (legal/policy): A-1 It is not a project that is non-compliant with international law, including international humanitarian law, and domestic law. A-2 It is not an autonomous weapon with lethal force that operates completely without human involvement.

Requirement B (technical) includes B-1 Clarification of Human Responsibility; B-2 Fostering Operator’s Appropriate Understanding; B-3 Ensuring Fairness; B-4 Ensuring Verifiability and Transparency; B-5 Ensuring Reliability and Validity; B-6 Ensuring Safety; and B-7 Compliance with International Law and Domestic Law.

Classification: if the output resulting from the AI function of the equipment does not affect functions possessing destructive capability, it is a low-risk AI-equipped system; otherwise it goes through legal/policy review and, if eligible, high-risk technical review. Reviews run before the project commences for legal/policy, and at milestones … such as before project commencement, before design approval, before testing commencement, and before operational commencement for technical review.

The English table of contents includes An Example of Confirming Whether Requirement B Is Fulfilled Using the RAI Toolkit. That names the US AIA / RAI Toolkit as an example method in an appendix. No sentence we cite makes completing that toolkit a Japanese certificate.

The Japanese 第１版 PDF is the signed original. Use it with the ATLA English edition, and remember the English file is a provisional translation.

Questions

Is the July 2024 policy a certification scheme for Japanese suppliers?

No cited page treats 防衛省ＡＩ活用推進基本方針 as a product certificate. The press conference calls it a compass. The later ATLA guideline is a guideline that implementers should comply with when planning and implementing research and development projects in the R&D slice of the equipment lifecycle.

Does Requirement A-2 ban all autonomy?

The sourced sentence is that the system is not an autonomous weapon with lethal force that operates completely without human involvement, including that it is operated within a responsible chain of human command and control. We do not extend that sentence to every unmanned system.