The Act on the Promotion of Research and Development and the Utilisation of AI-Related Technologies (人工知能関連技術の研究開発及び活用の推進に関する法律), commonly called the AI Promotion Act or AI Act (Japan), is Japan's comprehensive AI statute, enacted by the National Diet on 28 May 2025. It frames AI governance as innovation policy rather than consumer protection: it establishes an AI Strategic Headquarters within the Cabinet, places only soft-law "endeavour to cooperate" duties on businesses, and attaches no penalties for non-compliance. The Future of Privacy Forum has described it as an "innovation-first" framework.
Status and timeline
The Act was sponsored by the Cabinet Office under Prime Minister Shigeru Ishiba and enacted by the National Diet (House of Representatives and House of Councillors). A draft was published for public comment in February 2025, the Cabinet approved it in April 2025, and the Diet enacted it on 28 May 2025. Its principal provisions came into force on 4 June 2025, and the government announced the Act as being in full effect in November 2025 (Government of Japan, "Act on Promotion of R&D and Utilization of AI-Related Technology Now in Full Effect," November 2025). The Act includes a five-year review clause.
On July 16, 2026, the government revised the AI Basic Plan — the operational plan adopted under the Act — to require AI developers and operators to implement cybersecurity practices, risk assessments, and transparency standards, and to expand the role of the Japan AI Safety Institute (Source: substack.com). The revision moves the framework's expectations of businesses beyond the statute's original soft-law cooperation duties, though the Act's no-penalty structure is unchanged.
Scope and definitions
The Act covers the research, development, and utilisation of AI-related technologies in Japan. It applies to national and local government, universities, research institutions, businesses, and citizens, with duties differentiated by actor. Unlike the EU AI Act and the Korea AI Basic Act, it contains no risk-tier structure. It imposes no direct obligations on foreign providers beyond general cooperation expectations, and treats AI governance as innovation policy rather than as a consumer-protection statute.
Key provisions
The Act is structurally and philosophically different from the comprehensive AI laws of the EU, Korea, and Brazil. Its principal mechanism is an inter-ministerial coordination body operating through a national plan, backed by non-binding guidelines rather than enforceable rules.
AI Strategic Headquarters
The Act establishes the AI Strategic Headquarters (AI戦略本部, Art. 6–12) within the Cabinet, chaired by the Prime Minister, with all Cabinet ministers and appointed experts as members. Its mandate is to formulate and implement the AI Basic Plan, drafted and approved through Cabinet decision. The Headquarters coordinates across ministries, including METI, MIC, and MEXT, and with local governments. The first AI Basic Plan is expected in 2026.
AI Basic Plan
The AI Basic Plan is the operational core of the Act. It sets national R&D priorities; infrastructure goals covering compute, data centres, and datasets; human-resources development; international cooperation goals, especially with the G7 and the Global Partnership on AI; safety and reliability research priorities; and public-sector AI adoption targets.
Duties by actor
Businesses are placed under a duty to "endeavour to cooperate" with government measures (Art. 14), a soft-law formulation (努力義務, doryoku-gimu, "obligation to endeavour"), and to contribute to safe, secure, and beneficial AI use. The Act imposes no specific compliance obligations, audits, or impact assessments on businesses. Citizens and users are directed to understand the nature and limitations of AI and to engage responsibly (Art. 15).
Safety provisions and guidelines
The government is directed to establish guidelines, rather than binding rules, for safety, reliability, and transparency (Art. 16–17). METI's AI Guidelines for Business (v1.1 2025, with an update expected in 2026) function as the operative soft-law baseline; an earlier version, v1.0, was issued in 2023. The AI Safety Institute (AISI Japan), established in 2024 under the Information-technology Promotion Agency (IPA), continues under the Act with an expanded remit and conducts safety evaluations.
Governance and institutional structure
Implementation runs through a layered set of bodies. The AI Strategic Headquarters in the Cabinet sits at the highest level. METI (Ministry of Economy, Trade and Industry) is the principal business-facing regulator through its guidelines. MIC (Ministry of Internal Affairs and Communications) handles the telecommunications and privacy intersection, and MEXT (Ministry of Education, Culture, Sports, Science and Technology) handles R&D. AISI Japan, under the IPA, conducts safety evaluations. The Personal Information Protection Commission (PPC) regulates AI-adjacent personal-data matters via the APPI. Sectoral regulators retain their existing authority, including the FSA for finance and MHLW for healthcare.
Enforcement and penalties
The Act attaches no penalties, financial or otherwise, for non-compliance. Enforcement rests on cooperation, reputation, and government-procurement leverage. This contrasts with the EU AI Act's penalties of up to €35M or 7% of turnover, Korea's fines (KRW 30M per infraction), and Brazil's 2% turnover cap.
Comparison with other approaches
| Dimension | Japan AI Promotion Act | [[eu-ai-act | EU AI Act]] | [[south-korea-ai-basic-act | Korea AI Basic Act]] | US patchwork | [[china-generative-ai-interim-measures | China CAC framework]] |
|---|---|---|---|---|---|---|---|---|
| Philosophy | Innovation promotion; minimal restriction | Risk-based; fundamental rights | Balanced innovation + trust | Sectoral + state-level | Content control + sovereignty | |||
| Risk tiers | None | Four-tier + GPAI | High-impact + AI of national significance | None federally | Licensing-based | |||
| Obligations on business | "Endeavour to cooperate" | Specific per-tier obligations | Specific per-tier obligations | Varies | Licensing, content moderation | |||
| Penalties | None | Up to €35M / 7% turnover | KRW 30M / infraction | Varies | Licensing revocation + fines | |||
| Generative-AI labelling | Voluntary via guidelines | Mandatory | Mandatory | No federal rule | Mandatory | |||
| AISI status | Operational, referenced in Act | EU AI Office supersedes | Statutory | Non-statutory | N/A | |||
| Enforcement | Soft law; reputational + procurement | Hard law + AI Office + national authorities | MSIT + K-AISI | Varied | CAC |
Among major AI statutes, Japan's adopts the lightest compliance burden, coordinating governance from the Cabinet rather than through a lead regulator, and relying on cooperation and government-procurement signals in place of penalties. The Act extends Japan's earlier "agile governance" documents — the METI AI Guidelines for Business v1.0 (2023) and "Agile Governance in a Multistakeholder World" (2021) — with statutory backing while leaving their substance unchanged. The CSIS analysis "Japan's Agile AI Governance in Action: Fostering a Global Nexus Through Pluralistic Interoperability" situates the Act within this agile-governance tradition.
It is the second major Asian comprehensive AI statute after Korea's, but structurally opposite in design. Japanese officials cite the EU AI Act as the "paralysing regulation" model they aim to avoid, presenting the statute as a pro-innovation alternative. Commentators describe an emerging three-model Asian landscape: China's content-control and licensing approach, Korea's risk-tier and rights approach, and Japan's pro-innovation and soft-law approach, with Singapore, Taiwan, and ASEAN states positioning between them.
Reactions and critiques
Civil-society groups, including Japan Computer Access for Empowerment and privacy-law scholars at Keio and Waseda, argue that the "endeavour to cooperate" framing is insufficient for high-stakes AI deployment and that, without penalties, compliance is effectively optional.
A second line of critique concerns consumer protection. Unlike Korea's Act, Japan's statute provides no individual rights — no right to explanation, to contest, or to human review of AI-mediated decisions. Consumer groups argue that existing consumer-protection, product-liability, and APPI law cannot cover AI-specific harms.
The de facto enforcement lever is that government-facing AI providers must adhere to the guidelines. Observers note this works for enterprise AI sold to government but has little reach over consumer-facing generative AI.
Japan's lighter touch is increasingly out of step with the EU and Korea. The UK approach is closer, though the UK is moving toward a promised AI Bill. The G7 Hiroshima Process, which Japan led in 2023, assumed convergence toward binding frameworks; the AI Promotion Act signals that Japan will not follow that convergence domestically. The Hiroshima Code of Conduct (2023), an international voluntary framework Japan led drafting on, is substantively stricter than Japan's own domestic statute.
The statute's effectiveness is widely tied to the AI Basic Plan, whose specificity and ambition remain to be published, and to the built-in five-year review, which may force revisions if the approach proves inadequate. The Act's voluntary, reputational, and multistakeholder model will be observable through 2030, the five-year review window.
Relationships
- related: EU AI Act (Regulation 2024/1689) — the model Japan explicitly declines to follow
- related: South Korea AI Basic Act — Source Summary — regional contrast: both are Asian comprehensive AI laws, structurally opposite
- related: G7 Hiroshima Code of Conduct for Advanced AI (2023) — Japan led this international voluntary framework; domestic statute is substantively thinner
- related: Brazil AI Bill (PL 2338/2023) — Source Summary — contrast: Brazil moving toward rights-rich risk-tier model
- related: Canada AIDA (Bill C-27, Part 3) — Source Summary — contrast: Canada attempted EU-style law and failed; Japan enacted pro-innovation alternative
- related: Australia Voluntary AI Safety Standard — Source Summary — parallel voluntary-first approach
- related: Singapore Model AI Governance Framework for Generative AI (2024) — parallel soft-law, multistakeholder approach
- related: AI Sovereignty — Japan's choice to regulate differently is an explicit sovereignty claim
- related: America's AI Action Plan — shares pro-innovation framing with US Trump-era AI Action Plan
Sources
- AI Promotion Act source summary
- Future of Privacy Forum, "Understanding Japan's AI Promotion Act: An 'Innovation-First' Blueprint"
- Government of Japan, "Act on Promotion of R&D and Utilization of AI-Related Technology Now in Full Effect" (Nov 2025)
- CSIS, "Japan's Agile AI Governance in Action: Fostering a Global Nexus Through Pluralistic Interoperability"
- Bird & Bird, "Japan's AI Governance — Major Government Steps Since the AI Act" (2025)
- International Bar Association, "Japan's Emerging Framework for Responsible AI"
- IAPP, "Global AI Governance Law and Policy: Japan"