Japan often stands as a country associated with a strong image of AI, notably due to its pioneering breakthroughs in regard to robotics, automated vehicles and connected devices.
However, behind the look, Japan is below EU in terms of AI development, and Japan AI legal landscape is in an embryonic state as its first binding legislation in the form of the Japan Act on the Promotion of Research, Development and Utilization of Artificial Intelligence-Related Technologies (hereafter, the “Japan AI Act”) entered fully into force only on 1 September 2025.
To celebrate such landmark legislation, what’s better than analysing its provisions in comparison with its European counterpart, regulation (EU) 2024/1689 (hereafter, the “EU AI Act”) as in force since 1 August 2024, with the goal to enlighten both Japanese and European business operators engaging in cross-border AI-involving activities in Japan or in the EU.
Once we transcend the obvious difference in size between the EU AI Act (113 articles and 180 recitals) and the Japan AI Act (only 28 articles), we can notice a major similarity (1) followed by key differences (2), the reason of these being that, while both legislations enshrine as a principle the regulation of AI, the main rationale behind the Japan AI Act is to promote the deployment and use of AI whereas the EU AI Act is mainly intended to frame AI development in order to protect individuals’ rights and users’ security.
A similar approach vis-à-vis AI and its usage
A similar approach consisting in subjecting the largest number of AI stakeholders possible (A) under similar AI usage fundamentals safeguarded by a centralized authority (B) drives both legislations.
A) Similarity in the broad definition of AI and in the wide target scope of application
Firstly, under the Japan AI Act, “AI technologies” (AI being jinkou chinou in Japanese) are defined as any technology necessary to realize functions that artificially substitute for human cognitive, reasoning, and judgment capabilities, and any technology related to information processing systems that utilise such to process input information and output the results (art. 1).
As such, AI is defined broadly like in the EU AI Act which revolves around a wide definition
of “AI systems”. Yet, the EU AI Act pushes further its will to catch thoroughly any AI in its net, by categorizing AI based on a risk-based approach, but also by integrating the definition of General-Purpose AI (GPAI) based on the concepts of purpose and use (art. 3).
Secondly, the Japan AI Act encompasses all stakeholders: The State, local governments, R&D institutes, operators and citizens (art. 4 to 10). Yet, the main subject to these obligations is the State which has to implement and promote AI R&D, shared-usage facilities, trained and educated personnel, appropriate AI usage (art. 10 to 17), as well as a “Basic Plan” (in Japanese, kihon keikaku) containing policies and measures (art. 18) and set to be published by winter.
In the same way, the EU AI Act also tends to address a maximum of stakeholders, both public and private, the difference being that the main party concerned is the private business “operator” englobing AI providers, product manufacturers, deployers, importers and distributors (art. 3).
B) Similarity in the fundamental principles guiding AI usage and governance
Firstly, under the Japan AI Act, fundamental principles (in Japanese, kihon linen) that drive AI usage are the principles of appropriate usage, transparency, human-centric safety, economic effectiveness, and societal inclusiveness amongst others (art. 3).
These are very close to the core principles guiding AI usage under the EU AI Act which are: Human agency and oversight; technical robustness and safety; privacy and data governance; transparency; diversity, non-discrimination and fairness; and societal and environmental well-being and accountability (see recital 27 for example).
Secondly, under the Japan AI Act, based on the same European idea that a new one-stop authority will best answer the needs for a coherent legislative application, AI governance relies on one centralized one-stop authority, the AI strategy headquarters (in Japanese, senryaku honbu) established within the national government to promote AI and its usage pursuant to the above fundamental principles, and notably via the drafting of the Basic Plan (art. 19 to 24).
Similarly, the EU AI Act establishes the AI Office, an EU-centralized agency within the Commission which has the mandate to develop tools such as guidance and codes to promote AI usage compliant with fundamental principles, completed by monitoring and supervision tasks (art. 64).
However, contrary to its Japanese variant, it is endowed with the power to investigate and sanction violations, marking one of the key differences with the Japan AI Act, as a divergence in the main intention behind each legislation exists: Promotion of the deployment and use of AI in Japan, and protection in fine of the society from the risks resulting from the latter in the EU.
Key differences in the application behind a similar approach
Differences reside in the hard law mechanism opted by the EU versus the soft law mechanism chosen by Japan as regards both the nature of obligations (A) and the intensity of sanctions (B).
A) Difference in application regarding the nature of obligations
The Japan AI Act is limited to obligations of cooperation for its addressees, including operators (in Japanese, katsuyou jigyosha) defined as any entity seeking to develop or provide products or services utilizing AI, or using AI for business (art. 7). As such, stakeholders other than the State (in Japanese, kuni) only endeavour to cooperate with the latter for the promotion of AI and for the application of State measures, constituting best efforts obligations (art. 5 to 9).
Incidentally, the softness of these obligations is even more flagrant given that best efforts notably apply only when concrete measures under the Basic Plan are published, the latter’s timing being unknown.
In contrast, the EU AI Act bans AI when it poses unacceptable risk (art. 5), imposes ex-ante features for high-risk AI such as a management system, recordkeeping or data training (art. 9 to 15) to be assessed before launch in the EU Market (art. 16), and imposes transparency obligations for limited-risk AI (art. 50), not to mention duties specific to GPAI.
The hardness of these obligations, when paired with concrete set timeline, is even more striking. Indeed, provisions for prohibited AI systems have been effective and subject to fines from 2 February 2025, most of its other provisions will be fully enforceable (including provisions on high-risk and limited-risk AI systems) as of 2 August 2026 (art. 113), and the provisions for GPAI apply in between 2 August 2025 and 2 August 2027 depending on the obligations at stake (art. 111).
B) Difference in application regarding the intensity of sanctions
Despite the Japan AI Act mentioning the role of the State in investigating infringements and conducting countermeasures (art. 16), it is silent regarding sanctions, reports evoking eventually at best a name and shame system in the future. This is due to Japan primarily seeing this law as a tool to promote AI face to the awareness that Japan AI usage and culture as well as AI inbound investments are underdeveloped (Cabinet Office report).
However, sanctions are real in the EU as non-compliance will result on maximum fines ranging from EUR 7,5 million or 1% of worldwide annual turnover to EUR 35 million or 7% depending on the level of risk associated with the infringing AI (art. 99). This reflects the EU legislator’s stance in viewing AI mainly as a threat from which individuals’ rights are to be protected.
Now, similarities and differences put aside, and going forward, the race for Japan to become the “world’s most AI-international country” (AI White Paper 2024) is on. And given the fact that the Japan AI Act cites international cooperation as a potential trigger for the latter’s adaptation in the future (art. 3), the question remains if the Japan AI Act will evolve towards a completer and coercive legislation like the EU AI Act.
Disclaimer: This article has no vocation to advise in Japanese law but only to give an informative general comparative viewpoint between European and Japanese legislations.
