Public research body · Science policy · Japan (AISI secretariat housed at IPA; IPA lists its headquarters in Bunkyo City, Tokyo). · Est. 2024
The Japan AI Safety Institute (AISI) is a Japanese government research and policy-technical organization focused on AI safety evaluation methods, standards, and practical guidance for safer AI development, deployment, and use. AISI describes its mission as developing and promoting evaluation approaches and criteria for AI safety, and operating as a hub that coordinates across related government ministries and organizations while also collaborating internationally with comparable AI safety institutes.
AISI states that it was established at Japan’s AI strategy level (AI strategy meeting) and began operation on February 14, 2024, with its secretariat function placed at the Innovation Platform Agency (IPA). The institute’s programmatic work is organized around (1) frameworks/standards and guidance (including evaluation “perspective” guides), (2) technical evaluation and verification capabilities and environments, (3) domain-specific safety evaluation activities through working groups and demonstration efforts (e.g., healthcare and robotics), (4) AI security-related survey/reporting and guidance, (5) standardization and international coordination, and (6) educational/publicity and dissemination aligned to safe implementation. In its published outputs, AISI emphasizes concrete artifacts—guides, checklists, assessment perspectives, and companion “playbooks” (e.g., for incident response and red-teaming)—that are intended to be reused by developers/providers and operationalized by organizations seeking to evaluate AI systems systematically.
Publications are not yet indexed for this organization. Read them on aisi.go.jp ↗.
AISI released a safety evaluation perspectives guide for AI robotics through its robotics sub-working group, outlining perspectives to identify and evaluate risks in AI robot deployments.
AIセーフティに関する評価観点ガイド(第1.20版)の公開AISI published an updated version of its AI safety evaluation perspectives guide, reflecting changes in the AI safety evaluation environment and expanded/updated evaluation items.
米OpenAI社とMOCを締結しましたAISI announced a Memorandum of Cooperation with OpenAI to explore cooperation opportunities related to AI safety evaluation, benchmarking, and cybersecurity, including exchange of views and ongoing dialogue.
データ品質マネジメントガイドブック Data Quality Management GuidebookAISI published its Data Quality Management Guidebook (Version 1.02) and a companion data quality management checklist (Version 1.00), aimed at sustained assurance of data quality for AI safety and effective data-driven operation.
「AIシステムに対する既知の攻撃と影響」に関する更新（第２版）AISI updated its document summarizing known attacks on AI systems and their impacts, describing the update as a second edition based on recently presented academic conference content.
ヘルスケア領域におけるAIセーフティ評価観点ガイドの公開AISI released a healthcare-focused AI safety evaluation perspectives guide and companion practice examples oriented to real-world evaluation execution.