Japan chose guidance over law for AI businesses

The design

Japan's Ministry of Economy, Trade and Industry (METI) and Ministry of Internal Affairs and Communications (MIC) jointly issued the “AI Guidelines for Business Ver1.0” on 19 April 2024. Their joint press release states the two ministries integrated and updated three earlier documents — the 2017 AI R&D Guidelines, the 2019 AI Utilization Guidelines, and the 2022 Governance Guidelines for Implementation of AI Principles — into a single document, following a public comment period held from 20 January to 19 February 2024. Unlike South Korea's enacted AI Basic Act or the EU's AI Act, this is not legislation.

What the evidence says

The guidelines document itself states plainly what kind of instrument it is: the ministries “decided to draw up guidelines on the basis of the goal-based concept that would lead to the achievement of purposes through soft laws without any legally binding force that would encourage interested parties to make voluntary efforts to reduce societal risks in AI.” That is the document's own characterization, not a secondary summary. It lists reasons for choosing this form over statute: a time lag between how fast AI develops and how slowly legislation can be drafted, and concern that detailed rule-based obligations could inhibit innovation.

What it asks of people

Directed at “AI business operators” rather than consumers, the guidelines ask developers and users of AI to observe listed principles, including one squarely relevant to companion-style systems: a section on “paying attention to manipulations by AI on decision-makings and emotions” states that operators should “not develop, provide, or use AI systems and services with purpose of manipulating human decision making, recognition and emotion,” and should take countermeasures against “the risk of heavy dependence on AI.” Because the document is voluntary, it asks operators to self-assess against these principles rather than certify compliance to a regulator.

Privacy and safeguards

The guidelines address accessibility for “socially vulnerable people” and note the risks of disinformation from generative AI, but they do not create data-retention rules, a complaint mechanism, or penalties; enforcement is not a feature of soft law. The document frames itself as one input into an evolving policy landscape, stating the ministries “will continue to update the guidelines with necessary information,” so any description of its content should be dated to the retrieval point rather than treated as fixed.

• Has METI or MIC published a later version of the guidelines superseding Ver1.0 as retrieved here?
• Does a given companion AI product's design fall within the guidelines' warning against manipulating “decision making, recognition and emotion”?
• How do Japan's voluntary guidelines compare in practice with South Korea's or the EU's enacted transparency statutes for the same category of product?

Japan's choice of a jointly issued, voluntary document over a statute is itself the primary fact here, and the guidelines' own text, current as retrieved on 16 September 2026, is what defines their scope and limits.