AI Regulation Tracker  /  Regulator guidance

Japan's METI publishes an AI contract checklist for allocating liability, IP, and indemnity

METI released a template that helps businesses decide who bears the risk when an AI system produces a bad output. It is guidance, not law, but it offers a practical reference for AI development and procurement contracts in Japan.

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

Japan's Ministry of Economy, Trade and Industry published a "Checklist for AI Use/Development Contracts" on February 18, 2025, giving businesses a structured template for one of the hardest questions in any AI deal: who bears the cost when the technology gets something wrong. The document, titled AIの利用・開発に関する契約チェックリスト, is aimed squarely at the people who write and sign these agreements, namely in-house counsel, IT vendors, and procurement teams.

What the checklist covers

The checklist frames an AI contract as an exercise in allocating both risk and benefit between the provider of an AI system and the user of it. According to METI, it walks through four questions that AI agreements often leave vague. First, liability for erroneous output: if the model produces a wrong or harmful result, which party absorbs the consequence. Second, intellectual property: who owns the training data going in and the outputs coming out. Third, indemnity: how one party protects the other against third-party claims. Fourth, the scope of data use: how far the provider or user may go in using the data that passes through the system.

None of these are new legal concepts. What the checklist adds is a single reference that names each decision point explicitly, so parties negotiate from a shared list rather than discovering the gaps after a dispute.

The practical value shows up in the moments where AI contracts tend to fail. Liability for erroneous output is the clearest example. A general-purpose model can produce a plausible but wrong result, and a contract that is silent on the point leaves both sides arguing after the fact about whether the provider promised accuracy or the user assumed the risk of a probabilistic system. Naming the question up front forces a decision instead of a later fight. The same logic runs through the other three categories. Ownership of training data and outputs determines who can reuse what the system learns and produces. Indemnity decides who stands between the counterparty and a third-party claim, such as an intellectual-property complaint over training material. Data-use scope sets the outer limit on what a provider may do with information that flows through the service, a term that matters more as models are retrained on usage data.

Guidance, not law

This is the point most easily missed. The checklist is a practical template, not binding regulation. It creates no statutory duty and carries no penalty for ignoring it. METI's stated purpose is to make risk and benefit easier to allocate so that businesses adopt AI with more confidence, not to impose a compliance regime.

It also sits inside a larger soft-law structure. The checklist is a companion to Japan's AI Guidelines for Business (AI事業者ガイドライン), the voluntary framework jointly maintained by METI and the Ministry of Internal Affairs and Communications that sets governance expectations for AI developers, providers, and users. Read together, the guidelines describe how organizations should govern AI, and the checklist descends one level to the contract that formalizes the relationship between two parties.

What it does not do

The checklist does not decide any of the four questions for you. It does not dictate that the provider must accept liability for erroneous output, nor that the user must own the outputs. It surfaces the choice and leaves the allocation to negotiation. It is not enforceable against a party that drafts around it, and it does not override the terms the parties actually agree. Treating it as a rulebook would overstate its force; treating it as a drafting aid is the correct read.

Why a US reader should care

For a multinational, the value is comparative. Japan has chosen to keep AI contracting inside ordinary commercial negotiation, supported by a government checklist, rather than mandating specific terms. That is a useful counterpoint to the European Union's AI Act, which pushes obligations down the supply chain through contractual flow-downs, and to the US market, where vendors increasingly bolt AI-specific addenda onto master agreements. A company standardizing its AI master service agreements across regions can use the METI checklist as the Japan column in that crosswalk, confirming that its global template addresses the four points Japanese practice now expects parties to settle. Because it is guidance rather than law, it binds no US firm operating in Japan, but it signals the baseline local counterparties will bring to the table.

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.