The Pulse
Japan Adds AI-Agent Tools to Its 180,000-Staff Government Platform
Japan’s Digital Agency began rolling out AI-agent tools through its GENAI platform to about 180,000 government employees on October 20, 2026. The agency also published rules assigning responsibilities to users, administrators and developers
AI.info Team · Written with AI · fact-checked ·
Japan’s Digital Agency began rolling out AI-agent tools across the government’s GENAI platform on October 20, with about 180,000 government employees in scope. The rollout starts with Agent Chat, a tool for creating and running agents; a second environment, GENAI Workbench, is targeted for around December 2026. The announcement also sets out new rules for how staff and administrators should manage the systems.
Two agent environments, on different schedules
Agent Chat is designed to let employees create and run agents through a web chat interface. The agency’s standards describe support for tasks including document drafting, data analysis, file creation, code execution and the use or creation of reusable agent skills. Ministries and agencies are to receive the tool in stages, so the October announcement does not mean every employee already has access.
GENAI Workbench is intended for more involved work: building and running advanced agents, gathering information across multiple websites, analyzing data and producing documents or working prototypes. The standards describe it as a personal development environment that can carry out sequences of web retrieval, file operations and command execution. The agency says its aim is to make agent capabilities available beyond technical specialists, including to administrative staff in day-to-day work.
“Government employees need an environment in which they can create and run AI agents tailored to their duties and objectives and continuously improve them through practical use,” the Digital Agency writes in its announcement.
Rules put staff in charge of consequential decisions
The accompanying GENAI Standards divide responsibilities among users, administrators and developers. Employees must check and, where needed, correct an agent’s output before using it; agency managers decide which tools and models staff may use and set local controls. The Digital Agency’s development team is responsible for the platform’s design, evaluation, monitoring and ongoing improvement.
The document says human oversight must match the nature and potential impact of a task. Where work could affect people’s rights, safety or daily lives, agencies must provide for staff review, approval, intervention and the ability to stop use. The standards leave the final judgment and responsibility with government employees and their agencies, rather than treating an AI system as an accountable decision-maker.
Domestic data handling comes with a stated exception
The standards say GENAI will generally use Japan’s Government Cloud, with AI processing and related data storage in domestic data centers. Providers must contractually agree not to use government prompts, files or outputs to train or improve their models. The agency may still collect and handle logs, including input and output data, for specified operational, security and improvement purposes, subject to defined access and use limits.
The rules allow a limited exception when an advanced model is not available in a domestic region. In that case, the agency says contracts must prevent input and output data from being stored overseas after processing, and users and administrators must be told about the risks. The document therefore sets domestic handling as the default, not an unconditional guarantee that every model operation stays in Japan.
GENAI’s government-wide trial began in May
The agent rollout builds on a larger GENAI trial that began in May 2026 and targets approximately 180,000 staff across ministries and agencies. The Digital Agency says it will use the trial to assess effectiveness, safety, usability and operational issues before ministries move toward full-scale use from fiscal 2027. Its announcement describes the agent environments as a world-first effort at this scale, a claim attributed to the agency’s own research.
The agency plans to release Agent Chat and GENAI Workbench as open-source software around February 2027, allowing local governments and private organizations to use, modify and redistribute them. For now, the immediate change is narrower: Agent Chat is beginning a phased government rollout, while the advanced Workbench remains scheduled for around December.