AISI Reports on AI Safety Evaluation and Environment Development
Overview
AISI reported on the first half’s activities and plans for the second half and beyond of its Business Demonstration Working Group, as well as the work of the Task Force on the AI Safety Evaluation Environment. It shared the achievements of each SWG and plans for future verification and guide development. Anthropic and OpenAI also presented the latest developments. The article concludes with AISI’s plan to further step up activities toward realizing AI safety.
Key points
- The Business Demonstration Working Group shared its first half activity status and plans for the second half and beyond.
- Six SWGs reported on their field-specific achievements, verification work, and guide development.
- The Task Force on the Evaluation Environment shared its activities for fiscal year 2026.
Overview
The AISI Business Demonstration Working Group held a briefing to report on the activities of each sub-working group (SWG) during the first half of fiscal year 2026 and share information among stakeholders on plans for business demonstrations in the second half and subsequent fiscal years. Guest speakers from Anthropic and OpenAI also presented the latest developments.
At a separate general meeting, participants shared the activities for fiscal year 2026 of a task force in which AISI and private companies consider joint development and enhancement of the evaluation environment, along with an overview of evaluation tools released as open-source software.
Impact
AISI plans to further step up activities toward realizing AI safety, including through the task force.
Details
The Healthcare SWG will verify the usefulness of the “Guide to AI Safety Evaluation Perspectives in the Healthcare Domain,” published in April 2026, from the perspective of AI providers, gather opinions from medical institutions, and revise it. The Robotics SWG published the “Guide to Safety Evaluation Perspectives for AI Robotics,” which summarizes results from real-world equipment evaluations, in the first half. In the second half, it is verifying the usefulness of the evaluation perspectives and conducting demonstration experiments to examine the operational design domain (ODD) for AI robots.
The Data Quality SWG developed and published a checklist in the first half. It will verify its effectiveness and identify issues in actual operations, make improvements, and also assess metadata quality. The Conformity Assessment SWG is working with international standardization activities to consider a framework for conformity assessment systems for AI systems, including human-AI collaboration. In coordination with the standard-type BRiDGE project, it will prepare reports and guides.
Established this fiscal year, the Education SWG has organized risks related to the use of generative AI in primary and secondary education. It plans to continue exchanging views with experts and others, then compile use cases, risks, and evaluation perspectives. The Geo AI SWG is considering a common framework guideline and a public surveying manual for AI-assisted digital topographic maps, aiming to enable safe use of geospatial information in AI systems.
The AI Safety Evaluation Environment is open-source software that supports safety evaluations by AI system developers and providers. It was released in September 2025, and a task force with private companies was established in October 2025. The task force is considering functional enhancements and future direction in light of user feedback and technological developments.