AI Regulation Tracker  /  Regulator guidance

Japan's AI Guidelines for Business v1.2 add coverage of AI agents and physical AI

METI and MIC jointly issued Version 1.2 of Japan's core AI governance guideline on March 31, 2026, adding definitions, risks, and countermeasures for AI agents and physical AI. It stays voluntary soft law, but it is the compliance baseline Japanese firms and their vendors are expected to meet.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

Japan updated its central AI governance document on March 31, 2026, and the change matters beyond its own borders. The Ministry of Economy, Trade and Industry (METI) and the Ministry of Internal Affairs and Communications (MIC) jointly published Version 1.2 of the AI Guidelines for Business, known in Japanese as the AI事業者ガイドライン. The headline addition is coverage of AI agents and physical AI, two categories that most national AI frameworks had not yet addressed in a dedicated, operational way. Version 1.2 succeeds Version 1.1, which the same two ministries issued on March 28, 2025.

What Version 1.2 actually adds

The guideline is organized around three roles: AI developers, AI providers, and AI users. Each role carries its own set of voluntary governance expectations across the AI lifecycle, from design and training through deployment and monitoring. Version 1.2 extends this structure into two areas it did not previously treat in depth. For AI agents, systems that can plan and act toward goals with reduced human intervention, the guideline provides definitions, identifies the risks that agentic behavior introduces, and sets out countermeasures along with worked examples. It does the same for physical AI, meaning AI embedded in machines that act in the physical world, such as robotics and autonomous equipment.

That combination of definition, risk, countermeasure, and example is what gives the update practical weight. It moves the conversation from abstract principle to something a compliance team can map against a real product. Earlier versions of the guideline set out cross-cutting principles such as safety, fairness, transparency, and accountability, and asked each role to apply them across the lifecycle. Version 1.2 keeps that spine and threads the two new categories through it, so an organization can ask, for a specific agent or physical AI system, which risks arise and which controls the guideline expects. For agentic systems, that includes the harder questions of how much autonomy to grant, how to keep a human able to intervene, and how to trace what an agent did and why.

Soft law, not statute

A point that professionals should hold onto: this is guidance, not binding law. The AI Guidelines for Business impose no statutory penalty and create no direct legal cause of action for breach. They sit in Japan's soft-law layer, alongside the country's broader promotion-first approach to AI. What the guideline does carry is expectation. In practice, Japanese firms, their boards, and their counterparties treat the document as the reference standard for responsible AI conduct, which is why it operates as a de facto compliance baseline even without enforcement teeth.

So the guideline does not, by itself, expose a company to fines, and it does not convert into an EU-style risk-tiered regime. It also does not replace sector-specific rules that already bind a given industry, and it does not create new licensing or registration requirements for AI agents or physical AI. Reading it as binding law would overstate it. Reading it as optional would understate how domestic regulators, investors, and enterprise buyers actually use it. In Japan's promotion-first model, guidance like this shapes conduct through expectation and reputation rather than sanction, which is a distinction US professionals used to a rules-and-penalties structure should keep in view.

Why a US reader should care

For a US firm, the cross-border angle is concrete. Version 1.2 appears to be the first major national AI guideline to formally scope agentic AI governance, which makes it a citable reference point while other frameworks are still settling. Companies watching the European Union's AI Act general-purpose AI codes take shape, or waiting on US federal direction, now have a national-level document that names AI agents and physical AI and prescribes controls for them. Multinationals can point to it when building internal governance for agentic systems, and firms selling agentic or robotics products into Japan should expect Japanese partners to ask how their controls line up with it.

What to do with it

The measured response is to treat Version 1.2 as an input to existing governance rather than a new legal obligation. Identify whether your organization is a developer, a provider, a user, or more than one, then read the role-specific expectations and the new agent and physical AI material against your current practices. Where the guideline names a risk you have not documented, that is a gap worth closing, not because the law compels it, but because the expected standard in Japan now includes it.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.