Open-Source AI Safety Evaluation and Task Force Summit

The AI Safety Institute (AISI) is a public institution that develops tools for AI safety evaluation and promotes evaluation by developers and providers of AI systems. Its director is Akiko Murakami, and it belongs to the Information-technology Promotion Agency.

In September 2025, AISI released the AI Safety Evaluation Environment as open-source software (OSS), providing a foundation for conducting AI safety evaluations.

Beginning in October 2025, AISI and several private companies organized the AI Safety Evaluation Environment Task Force and have been discussing the environment's direction and proposals for functional enhancements based on user feedback and technological trends.

With the evaluation environment released as OSS, evaluation tools were made available for developers and providers of AI systems to conduct AI safety evaluations.

Through ongoing discussions by private companies and AISI, proposals for the evaluation environment's future direction and functional enhancements are being considered based on user feedback and technological trends.

The inclusion of AI agents in multiple discussions at the general meeting showed that the technology is attracting interest not only from an AI safety perspective but also from an AI business perspective. AISI aims to achieve AI safety by revitalizing activities, including the task force, from fiscal 2026 onward.

The general meeting was held with both general and online participants, aiming to explain the outlines of AISI, the evaluation environment, and the task force to participants without prior knowledge and increase their interest. The first half included an introduction to AISI's activities, explanations and demonstrations of the evaluation environment's functions, a report on the task force's activities, and introductions of its members.

The second half featured three discussions with task force members serving as panelists. The themes were AI red teaming and governance, AI agents and agentic AI, and AI business in 2026, with exchanges of views from both AI safety and business perspectives.

The event took place from 1:30 p.m. to 5:10 p.m. on February 26, 2026, at a conference room of Mitsubishi Research Institute, Inc. The 10 presenting companies were SB Intuitions, NTT, NTT DATA Group, NTT DOCOMO BUSINESS, Citadel AI, NEC, Nomura Research Institute, Fujitsu, Preferred Networks, and Ridge-i.