Contents
According to a November 11, 2026 report by Kyodo News, Japan’s government is considering requiring businesses handling highly capable AI to ensure safety and introducing new penalties. The proposal would enable authorities to act against businesses that refuse to cooperate in addressing serious risks, with an outline of legislative amendments expected by year-end.
Key Points#
- The government is considering legally binding orders and penalties for businesses that fail to cooperate in addressing serious risks posed by highly capable AI.
- Which businesses would be covered—including researchers, model developers, and service providers—has yet to be decided.
- Penalties are expected to focus on cases involving highly capable AI, but the report provides neither a definition nor details of the penalties.
- Companies should know who controls their AI systems and who can respond to government safety inquiries.
What Japan’s Government Is Considering#
According to Kyodo News, the government is reviewing its AI framework with a view to imposing safety obligations on businesses and introducing penalties. This approach was confirmed at a closed-door working group discussing AI legislation.
The reported enforcement mechanism has two stages. Authorities would issue legally binding orders as needed to businesses that fail to cooperate in addressing problems, then impose penalties for violating those orders. The aim is to enable the government to respond effectively to serious threats, such as misuse in cyberattacks or AI behavior that escapes intended controls.
This proposal remains under consideration. As of November 12, the supplied report does not confirm that legislative amendments have passed or that new obligations have taken effect.
Businesses That Could Be Covered#
The scope remains under discussion. Various businesses are involved in developing, deploying, and operating AI, and the government has yet to decide where legal responsibility should lie.
The report identifies research, foundation model development, and service provision to users as activities that could be covered. Companies building applications using other companies’ models should therefore avoid assuming that only model developers would face obligations.
The treatment of overseas businesses also remains unresolved. Kyodo News notes that, if major U.S. technology companies are included, ensuring effective protection for users in Japan would be a challenge. The report does not explain how orders would be communicated to or enforced against overseas businesses.
Why Legally Binding Orders Matter#
The proposed review addresses a specific problem: what authorities can do when a business refuses to respond to serious AI safety concerns. Requests for cooperation have limited effect if recipients can ignore them without consequences.
Legally binding orders would give the government’s response enforcement power. Their effectiveness, however, depends on whether they reach businesses with the authority and technical capacity to address the risks. A service provider may control user access while a separate model developer controls changes to the underlying model.
Kyodo News also references reports of unintended unauthorized access by advanced AI under development in the United States and suspected AI misuse in cyberattacks. These illustrate the concerns behind the review; they do not establish that AI caused any particular attack.
What Companies Can Do Now#
Preparation should begin by clarifying operational responsibilities. Companies can take four practical steps before the proposal is finalized.
1. Inventory the AI models and services they develop, provide, or use, including the external companies involved. 2. Designate one person responsible for receiving safety inquiries and coordinating responses. 3. Document who can restrict access, suspend services, or change permissions in a serious incident. 4. Review supplier contracts for provisions on incident notification, technical cooperation, and access to records needed to investigate problems.
These are preparatory measures, not obligations confirmed by the report. They are useful immediately, however. Companies should understand which decisions they can make themselves and which require supplier cooperation.
What Remains Undecided#
Four details will determine the proposal’s scope: the definition of highly capable AI, the businesses covered, the safety obligations imposed, and the penalties for violating orders. The supplied report specifies no capability thresholds, fine amounts, or effective date.
The government plans to present the full review, including the proposed legislative amendments, by the end of 2026. This is a deadline for setting out policy direction, not a confirmed enforcement start date. Companies should review the published text before treating the reported proposal as a compliance obligation.
Next Step#
Create a one-page inventory of your company’s AI systems, identifying who is responsible for managing each system and who can shut it down if a serious safety problem arises.
Source: https://www.47news.jp/15215887.html
Contents
