AI Regulation Tracker  /  Signal

Japan's METI Issues Guidance on How Civil Liability Applies to AI Use

On April 9, 2026, Japan's Ministry of Economy, Trade and Industry published a guide explaining how existing civil liability rules apply when AI causes harm. It sorts AI into an assistive type and a reliance type and works through the analysis for each. This is guidance, not a new law, and it changes no legal duty.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 16, 2026  ·  Last updated July 16, 2026

On April 9, 2026, METI published a guide on how civil liability is likely to work when AI is involved in causing harm. The Japanese title translates to the "Guide on the Interpretation and Application of Civil Liability in AI Utilization," and METI marks it as Version 1.0. It does not write new law. It takes the law Japan already has, mainly tort liability under the Civil Code and product liability, and explains how those rules would probably apply when someone is injured in connection with AI use. The reason METI gave for doing this is that AI can be opaque and can act with a degree of autonomy, and there has not been much case law or settled thinking to tell companies where responsibility would land.

The two types of AI the guide draws a line between

The center of the guide is a split. METI sorts AI into two categories based on how it is actually used, and then runs a separate liability analysis for each.

The first is assistive or supportive AI. In METI's words, "Assistive/Supportive AI refers to a type of AI that is used solely to assist or support the judgment of AI users, and is designed so that human judgment and action are ultimately involved." A person is still in the loop, still making the call, and the tool is helping them get there.

The second is reliance or substitutive AI. Again in the guide's own framing, "Reliance/Substitutive AI refers to a type of AI that, unlike Assistive/Supportive AI, is intended to function as a substitute for human judgment and action." Here the system is meant to stand in for the human decision rather than support it, and the guide ties that category to expectations about necessity and about the accuracy and safety the system needs to reach.

Why the line matters is simple. The more a deployment leans on the machine to replace human judgment, the more the analysis of who is responsible when it goes wrong shifts. Framing the tool as a helper that a person supervises is a different liability posture than handing it the decision.

How the guide sets up responsibility

The guide works from three parties: the developer or provider of the AI, the user who deploys it, and the third party who gets hurt. It centers on the default rules, tort and product liability, that apply when a contract has not already sorted out who bears what. In other words, it describes the floor that exists whether or not two companies have papered their deal.

METI does not stop at theory. The guide works through a set of hypothetical cases across industries, including logistics, legal technology, creative work, real estate review, manufacturing, delivery robots, and customer service, and walks the liability reasoning through each. It also addresses the position that a company that did its homework should be in a better spot. Where a party has looked at the risks, built systems that reflect the principles in Japan's AI business guidelines, and put reasonable countermeasures in place that match the level of risk involved, the guide treats that as a factor that can reduce the chance of being found to have breached a duty to avoid harm. That is not a safe harbor and METI does not call it one. It is a description of how ordinary negligence reasoning would weigh reasonable care.

The guide sits alongside Japan's broader AI framework. It complements the AI Promotion Act that took effect in 2025 and the AI business guidelines that METI and the Ministry of Internal Affairs and Communications maintain. None of those pieces turns this guide into a binding rule. They are the context it reads against.

What this is, and what it is not

I want to be careful here, because a government ministry putting its name on a document about liability can read bigger than it is.

This is a guide. It interprets existing law. It does not enact a statute, issue a regulation, or set a standard anyone has to meet. It does not create a new cause of action, it does not flip the burden of proof onto developers, and it does not impose a filing or disclosure duty. Nobody's legal obligations in Japan changed on April 9 because this came out. What changed is that companies now have METI's considered read on how a court is likely to think about AI-related harm.

That read has real value even though it binds no one. Courts are not required to follow it, but a well-reasoned interpretation from the ministry that oversees this area is the kind of thing parties, counsel, and eventually judges tend to reach for when the case law is thin. METI's own goal is predictability: it wants developers, providers, and users to be able to see the shape of their exposure and to use that when they manage risk and write contracts. Treat it as a strong signal about direction, not as a wind you are legally required to turn into.

What this means for US firms operating in or with Japan

If your company builds, sells, or runs AI that touches Japan, this guide is worth reading closely even though it is not your regulator speaking. It is the clearest current statement of how Japanese civil liability is likely to fall on AI-related harm, and that exposure is real for a US firm with customers, operations, or partners there.

The practical thread is the assistive versus reliance distinction. How you position and actually operate a deployment matters. A tool sold and used as an aid that a qualified person supervises sits in a different liability posture than one marketed to replace the human decision, and the guide's reasoning rewards keeping a human genuinely in control where the stakes are high. The other thread is documented reasonable care. The guide treats risk assessment, governance aligned with Japan's AI guidelines, and countermeasures sized to the risk as things that can help on the question of whether a duty was breached. That lines up with how careful US operators already think about AI risk, which makes the guide useful as a benchmark rather than a shock.

None of this is a US legal requirement. If you change a contract term, an allocation of liability, or a control because of this, base that decision on Japanese counsel and your own applicable law, not on a summary of a guide. But if you have exposure in Japan, this is a document your legal and risk teams should have in front of them.

What to do now

Read the actual instrument, not the headline. This is METI interpreting existing liability law, not Japan passing a new AI liability statute. Map where your AI touches Japan, since customers, operations, vendors, and partners there are what put you inside the guide's world. Be honest about which of your deployments are assistive and which are reliance, because the guide reasons about them differently and how you run a tool can matter as much as how you label it. Keep the evidence of reasonable care that the guide treats as relevant: risk assessments, governance tied to Japan's AI guidelines, and countermeasures scaled to the risk. And do not treat this as a rule. Any change to a contract or control should rest on Japanese legal advice and your applicable standards, not on this research note.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any law or guidance applies to your situation with qualified counsel in the relevant jurisdiction.