Public research body · Science policy · Est. 2024
The Japan AI Safety Institute (AISI) is a Japanese government-supported organization focused on improving the safety evaluation of AI systems. Its mission centers on developing and promoting assessment methods, standards, and guidance that enable AI developers, providers, and users to evaluate and manage AI safety risks more reliably.
AISI describes its core work as examining AI safety assessment techniques and criteria, supporting the creation of guidelines, and coordinating international collaboration with similar bodies in other countries (including the AI Safety Institute networks in the UK and the US). AISI is hosted through Japan’s Information-technology Promotion Agency (IPA) as a secretariat, with a formal launch date of February 14, 2024.
No curated programs yet. Check the careers page ↗.
Publications are not yet indexed for this organization. Read them on aisi.go.jp ↗.
A monthly policy roundup page that, among other content, references AI safety-related international and domestic policy context as part of AISI’s information-sharing activities.
AIロボティクスに関するセーフティ評価観点ガイドの公開AISI’s robotics sub-working group released an “AI Safety Evaluation Perspective Guide for Robotics,” organizing AI-robot risks into categories (physical, psychological, and social) and providing evaluation perspective items, including example use cases referenced by AISI’s evaluation guide.
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 “Guide to Evaluation Perspectives on AI Safety” (Version 1.20), describing expanded evaluation perspectives and example items, including an added perspective for agent systems that addresses “observation and control.”
「JAI-Trust：日本の生成AIの安全性とセキュリティーのベンチマーク構築プロジェクト」を開催しました。AISI hosted the “JAI-Trust” project event focused on building safety and security benchmarking foundations for generative AI in Japan, discussing directions and issues for objective evaluation and standardization.
データ品質マネジメントガイドブック（Data Quality Management Guidebook）公開AISI released “Data Quality Management Guidebook (Version 1.02)” and associated checklist materials (Version 1.00), positioned as framework guidance for data quality management in support of appropriate AI realization.
AIセーフティ評価環境 検討タスクフォース総会を開催しました。AISI held a meeting for the “AI Safety Evaluation Environment Review Task Force,” describing the OSS “AI Safety Evaluation Environment,” its development and next directions, and reporting on 2025 activities and planned enhancements. The announcement states that about 100 people attended (in-person plus online).