Japan’s Communications Ministry Starts Revising Its AI Business Guidelines to Cover AI Agents and Physical AI
Japan’s Ministry of Internal Affairs and Communications on October 30 began work on revising the AI Guidelines for Business, the voluntary rulebook it maintains with the Ministry of Economy, Trade and Industry, so that it says more about autonomous AI agents and physical AI, Nikkei reported. The two ministries aim to publish an updated version around late March 2027.
The discussion opened at the 32nd meeting of the ministry’s AI governance study group, held online. According to the meeting materials, the secretariat said the current version 1.2, released on March 31, 2026, added definitions of AI agents and physical AI but covered them only at a basic level, attached to sections written for conventional and generative AI. It proposed a table setting out, across six categories, what developers, providers and users should each do.
Under that draft, developers would limit the data a model can reach and build in circuit breakers that halt abnormal behaviour, providers would keep tool and API permissions to a minimum and monitor services as a whole, and users would regularly check that agents stay within their permissions and spell out how to stop them when something goes wrong. Human approval steps, management of non-human accounts and credentials, and the reliability of data that agents store in memory are also on the list. For physical AI such as robots, the secretariat listed hardware safety devices that work independently of software, emergency stop buttons and testing in virtual environments before deployment.
What it means in practice
The guidelines carry no penalties, but they shape how Japanese companies write internal AI rules, and a ministry survey put awareness of them at 81% and use at 46%. That gap matters because much of the new checklist falls on users, from reviewing logs to training staff, which may be hard for small firms and local governments. The Digital Agency told the meeting it may fold new agent provisions into its own rules for government AI use, which already underpin steps such as its move to open AI agents in Gennai to 180,000 government staff under safety rules. The revision also runs alongside the government’s review of Japan’s AI law, which is weighing penalties.
What to watch next is how the study group answers the secretariat’s question about whether any measures are too difficult to carry out, whether the final text separates minimum and recommended steps by company size, and how closely it tracks foreign frameworks such as the NIST AI risk management framework.