AI Regulation Tracker  /  Legislation in force

Japan's AI Promotion Act carries no fines but creates a public naming risk for non-cooperating firms

Japan's first dedicated AI statute imposes no monetary penalties, yet it hands the government a naming power over non-cooperating firms. Generative-AI providers are expected to publish a public acceptance statement and explain their training-data and IP practices.

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

Japan's AI Promotion Act (Act No. 53 of 2025) took full effect in September 2025, and its most quoted feature is what it lacks. The Act on Promotion of Research, Development and Utilization of AI-Related Technologies, known as the AI Promotion Act and formally Act No. 53 of 2025, carries no monetary penalties. It was enacted on May 28, 2025. For general counsel scanning the global regulatory map, that single line has done a lot of damage, because "no penalties" is being read as "no obligations." That reading is wrong.

What the Act actually does

The law establishes the AI Strategy Headquarters, a body chaired by the Prime Minister and coordinated through the Cabinet Office, to set national AI policy. On the private sector, it lands lightly in statutory terms. The sole direct obligation on businesses is to "endeavor to cooperate" with the measures the government adopts. That is a soft-sounding duty, and there is no fine attached to breaching it.

What makes the Act bind in practice is the enforcement channel. Where the government judges that a firm is not cooperating, it can issue guidance, send information requests, and, critically, publicly disclose the names of non-compliers. In a market where reputation and government relations carry real commercial weight, public disclosure functions as the sanction. The statute swaps financial penalties for a naming power.

That design is deliberate. Japan built a promotion-first framework meant to encourage AI development rather than deter it, and it kept the compliance apparatus deliberately light so as not to chill investment. But light-touch is not no-touch. The information-request power gives regulators a lawful route to demand documentation of how a provider trains and governs its systems, and the disclosure power gives them a public consequence for firms that stonewall. For a compliance function, that combination is the operative risk, not any line-item fine.

The disclosure expectations providers keep missing

The government pairs the cooperation duty with specific expectations for generative-AI providers. According to the government's own materials and the firm analyses tracking the Act, providers are expected to publish a website acceptance statement and to operate on a comply-or-explain basis across three points: disclose an overview of the types of training data used; maintain internal IP-protection controls and publish a summary of that policy; and, when a third party asserts a legal-rights request, confirm whether specific webpages were used in training a given model.

A caution on sourcing. The precise three-point comply-or-explain expectation is drawn from the government's stated position and from firm analyses of the Act, not from a single identified statutory clause. Readers should treat it as a government-stated expectation that providers are expected to meet, rather than a numbered provision they can cite in a filing. The load-bearing statutory facts, the absence of penalties, the "endeavor to cooperate" duty, and the enforcement-by-disclosure design, are confirmed across the official English summary and multiple firm analyses.

What it does not do

The Act does not create a licensing regime, a risk-tier classification like the EU AI Act, or a private right of action. It does not fine anyone. It does not, on its face, prohibit any particular model or use case. It is a promotion-and-coordination statute first, and a light-touch obligation second. Firms should not overclaim its reach any more than they should dismiss it.

The cross-border angle

For a US reader, the practical risk is misreading. A US or EU firm operating in Japan that treats the missing penalty as the end of the analysis will skip the disclosure preparation the government expects, and will be exposed to a public naming it did not see coming. Japan has chosen reputational enforcement over financial enforcement, which is a different instrument, not a weaker intention. The firms that handle this well will have their acceptance statement and their training-data and IP-protection documentation ready before any information request lands.

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.