AI Safety Institute

What AI Safety Institute does

AI Safety Institute (J-AISI) is a Japanese government institute established on February 14, 2024 and operated by the Information-technology Promotion Agency (IPA) under the Cabinet Office. Its mandate is to develop and promote AI safety evaluation methodologies, frameworks, and tools in response to growing international interest in AI safety, with a portfolio that includes the Guide to Evaluation Perspectives on AI Safety (10 perspectives, v1.10), the Guide to Red Teaming Methodology (v1.10), the Data Quality Management Guidebook (v1.0), the Healthcare AI Safety Evaluation Guide, the Chief AI Officer Guides, and the AI Incident Response Approach Book (AI-IRS). Its core technical product is aisev, an open-source (Apache 2.0) Python-based AI Safety Evaluation Tool released on GitHub in September 2025 that supports automated red teaming and quantitative/qualitative scoring against the institute's 10 evaluation perspectives.

J-AISI serves AI developers and providers, government ministries and agencies, private-sector organizations implementing AI systems, and healthcare AI developers, providing standards, frameworks, and conformity assessment guidance at no cost. Distribution is entirely non-commercial via the aisi.go.jp website, GitHub, and direct embedding of guidance into AI assistants such as ChatGPT, Claude, and Gemini through markdown-formatted documentation. The institute coordinates internationally through the AISI International Network (US, UK, Singapore, Australia, Korea, Europe/INESIA), has signed Memoranda of Cooperation with OpenAI and Anthropic, and conducts joint testing exercises with Singapore and Korea AI Safety Institutes, including a 660-run study on autonomous AI agent data leakage. Domestically, the J-AISI Partnership and Business Demonstration Working Groups (Healthcare, Robotics, Data Quality, Conformity Assessment) convene private-sector stakeholders including NTT, Fujitsu, NEC, Nomura Research Institute, Citadel AI, Preferred Networks, SB Intuitions, NTT Data, and Ridge-i to align Japanese AI safety practice with international frameworks such as the NIST AI RMF.

Name

AI Safety Institute

Legal name

Information-technology Promotion Agency, Japan (IPA)

Website

https://aisi.go.jp

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

1–10 employees

Short description

AI Safety Institute (J-AISI) is a Japanese government institute under the Cabinet Office that develops open-source evaluation tools, frameworks, and guidelines for AI safety, serving AI developers, government agencies, and private-sector organizations in Japan and globally.

Ownership category

akta.pro rank

Product category

AI Safety Evaluation Software

akta.pro primary industry

Regulatory Readiness & Audit Automation (e.g., EU AI Act, NIST AI RMF, ISO/IEC 42001) (HDAAAMAE)

akta.pro secondary industries

Safety & Alignment Evaluation (red-teaming, harmful capability testing) (HDAAAMAL), AI Observability, Monitoring & Evaluation Platforms (Drift, Quality, Safety) (HDAEANAF), Audit, Explainability & Accountability Tooling (traceability, reporting) (HDAAAKAL), Documentation & Reporting Automation (Model Cards, Datasheets, Impact Assessments) (HDAAAMAJ)

Keywords

• AI safety evaluation
• Red teaming methodology
• AI governance frameworks
• Conformity assessment
• AI incident response

Headquarters

HQ city

Akasaka

HQ country

Japan

HQ region

Asia

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Operations, Marketing or Sales

Distribution channels4 records

Marketing channels7 records

Core offering

The Japan AI Safety Institute (AISI/J-AISI) develops and publishes AI safety evaluation methodologies, frameworks, guidelines, and open-source tools for assessing trustworthy AI systems. It operates as a government-backed institute providing standardized evaluation perspectives, red teaming methodologies, data quality frameworks, incident response approaches, and a free open-source AI Safety Evaluation Tool (aisev) available on GitHub.

Product overview

Japan AI Safety Institute (AISI) operates as a government-supported institute focused on AI safety evaluation frameworks and methodologies rather than a commercial product company. Its portfolio consists of guidance documents, frameworks, and open-source tools: the core AI Safety Evaluation Tool (aisev) provides automated evaluation capabilities; foundational frameworks including the Evaluation Perspectives Guide (10 perspectives), Red Teaming Methodology Guide, and AI Incident Response Approach Book establish evaluation standards; domain-specific guides address healthcare AI safety; supporting resources include Data Quality Management Guidebook, Chief AI Officer Guides, and translated materials including NIST AI RMF crosswalks. International collaboration through partnerships with US, UK, Singapore, Korea, Australia, and European AI safety institutes supports global AI governance alignment.

Differentiator

Problem solved

Functional benefit

Products and services

• AI Safety Evaluation Tool (aisev) Open-source software tool for conducting AI safety evaluations, providing automated red teaming and quantitative scoring based on AISI's evaluation perspective framework. Released on GitHub under Apache 2.0 license and available for AI developers and providers worldwide.
• Guide to Evaluation Perspectives on AI Safety Fundamental guide presenting 10 evaluation perspectives for AI safety assessment covering harmful information control, misinformation prevention, fairness, privacy protection, security, explainability, robustness, data quality, and verifiability.
• Guide to Red Teaming Methodology on AI Safety Methodology guide for evaluating AI system risk mitigation measures from attacker perspectives, including practical examples with RAG-implemented AI systems. Version 1.10 published March 2025.
• AI Incident Response Approach Book (AI-IRS)

Quantifiable outcome

• AI Safety Evaluation Tool downloaded from GitHub for conducting automated red teaming and safety evaluations
• +1 more outcomes

Segments5 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability7 records

Feature9 records

Partnerships

Eleven partnerships are on record, tiered core and flagship.

• OpenAIcoreStrategic or Co-development Partner · 29 May 2026Memorandum of Cooperation (MOC) signed with OpenAI for collaboration on AI safety. This partnership enables joint activities related to AI safety evaluation and research between J-AISI and one of the leading AI developers globally.
• AnthropiccoreStrategic or Co-development Partner · 29 October 2025Memorandum of Cooperation (MOC) signed with Anthropic for collaboration on AI safety evaluation and research. Enables knowledge sharing and joint activities on AI safety with another major AI developer.
• Korea AI Safety Institute (KISA)coreStrategic or Co-development PartnerJoint testing exercise conducted with Korea AI Safety Institute examining how autonomous AI agents handle sensitive data during routine workplace tasks. Designed 11 scenarios across customer service, enterprise productivity, and personal productivity agent archetypes with 660 total test runs.
• Singapore AI Safety InstitutecoreStrategic or Co-development PartnerJoint testing exercise with Singapore AI Safety Institute on AI agent data leakage risks. Findings published showing no model was consistently correct and safe across 660 runs, with Singapore's highest-performing model achieving full safety in 57% of runs and full correctness and safety in 40%.
• US AI Safety Institute (AISI)coreStrategic or Co-development PartnerActive collaboration including participation in Stanford AI Symposium and panel discussions. Exchange of AI safety methodologies and coordination on international AI safety frameworks.
• UK AI Security Institute (formerly AISI)coreStrategic or Co-development PartnerActive collaboration through AI safety summits and international forums. UK institute participated in AI Seoul Summit and Hiroshima AI Process discussions.
• AISI International NetworkcoreStrategic or Co-development PartnerInternational network of AI safety institutes including counterparts from US, UK, Singapore, Australia, Korea, and Europe (INESIA in France). Network published joint Mission Statement in November 2024 and coordinates on international AI safety research and frameworks.
• AISI Partnership (J-AISI Partnership)coreStrategic or Co-development PartnerCollaborative framework established with support from relevant ministries and agencies, connecting domestic organizations for promoting AI safety initiatives. Partnership agreement formalized in August 2024, with participating organizations collaborating through working groups and sharing AI safety best practices.
• NIST (US National Institute of Standards and Technology)flagshipStrategic or Co-development PartnerCrosswalk collaboration producing mappings between Japan's AI Business Operator Guidelines and NIST AI Risk Management Framework (AI RMF). Japanese translation of NIST AI RMF and Playbook published with NIST permission. Crosswalk 1 and Crosswalk 2 completed.
• AI Safety Evaluation Environment Task Force MemberscoreStrategic or Co-development PartnerJoint task force with private sector companies including SB Intuitions, NTT, NTT Data Group, Citadel AI, NEC, Nomura Research Institute, Fujitsu, Preferred Networks, and Ridge-i. Task force discusses future direction and functionality enhancements for AI safety evaluation tools.
• AIセーフティ評価環境 検討タスクフォース参加企業coreStrategic or Co-development PartnerMultiple Japanese companies participating in the AI Safety Evaluation Environment Task Force to improve evaluation tools based on user feedback and technology trends. Companies contribute expertise in AI business for functionality enhancement discussions.

Recent moves10 records

Expansion highlights6 records

Direct peers

• Korea AI Safety Institute (KISA): Korea's national AI safety body, partnering with J-AISI on the 660-run joint agent testing exercise. Both institutes operate within the AISI International Network with comparable government-backed mandates for AI safety evaluation.
• US AI Safety Institute (under NIST): US government AI safety body housed within NIST, conducting safety testing of frontier models and developing AI evaluation frameworks. Direct peer with identical mandate for pre-deployment model evaluation and international network coordination.
• Singapore AI Safety Institute (under IMDA): Singapore's AI safety body focused on AI safety evaluation and international collaboration. J-AISI partnered with Singapore AISI on joint AI agent testing (660 runs) and both operate within the AISI International Network.
• UK AI Security Institute: UK government AI safety body conducting pre-deployment evaluations of frontier models and publishing evaluation methodologies. Closest direct peer — both are government-backed safety institutes with comparable mandates for AI safety evaluation and international network membership.
• Citadel AI: Tokyo-based AI safety/evaluation company providing AI testing, monitoring, and compliance tools. Member of J-AISI's AI Safety Evaluation Environment Task Force and operates in the same Japan-centric AI evaluation market as J-AISI.

Emerging players

• MLCommons: Industry consortium developing AI benchmarks and safety evaluation standards (e.g., MLPerf, AI Safety Benchmark). Operates in adjacent space of AI evaluation standards with potential overlap with J-AISI's framework development.
• Robust Intelligence (Cisco): AI firewall and safety testing platform acquired by Cisco, providing automated AI model testing and protection. Operates as a private-sector comparator to J-AISI's open-source evaluation tooling with stronger engineering resources but narrower government mandate.

Broad incumbents

• Anthropic: Frontier AI lab with industry-leading focus on AI safety research, including Constitutional AI and mechanistic interpretability. MOC partner with J-AISI (Oct 2025) and subject of J-AISI safety evaluations.
• OpenAI: Leading frontier AI lab with active AI safety research and red teaming programs. MOC partner with J-AISI (May 2026) and subject of J-AISI safety evaluations — also a major reference point for J-AISI's framework development.
• NIST (National Institute of Standards and Technology): US government standards body that developed the AI Risk Management Framework (AI RMF) which J-AISI has formally crosswalked with its own guidelines. NIST is the de facto global reference for AI standards and a direct collaborator on J-AISI's framework interoperability work.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights8 records

Customer concentration

• GitHubgithub.com/Japan-AISI
• LinkedInlinkedin.com/company/japan-aisi

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles4 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about AI Safety Institute