Japan’s AI Governance: Flexibility & Good Design
Japan became the second major Asia-Pacific economy to enact AI-specific legislation last summer with the Japan Act on Promotion of Research, Development and Utilisation of Artificial Intelligence-Related Technologies (Act No.53 of 2025).
Japan’s sober strategy should be a welcome approach for companies developing, providing, and using AI, especially given the sensational reversals in the EU AI Act‘s framework, patchwork US State AI laws, and the penalties, registration requirements, and broad compliance regimes of other developed economies.
Three Layers, One Strategy
Japan has deliberately constructed a layered governance architecture that combines its new promotional statute, detailed voluntary guidelines, and the reliable enforcement power of existing law.
The result is a system thoughtfully designed to be:
- Light enough to attract and support investment and research activity;
- Flexible, with clear foundational bodies and principles to be elevated into place as international norms and domestic risks crystallize;
- Predictable, allowing for substantial adaptive lead time and reserving hard legal consequences only for areas of well-established and confirmed law.
The Framework Statute
The potential impact of AI on developed economies requires the highest level of government attention. However, the actual shape of that impact is far from a settled outcome.
The definition of AI-related technology in the AI Promotion Act reflects this broad impact: technology that substitutes for human cognitive reasoning and judgment abilities, as well as information processing systems that use such technology.
Recognizing this dynamic, the Act establishes the necessary top-level architecture without prematurely prescribing rules. It creates an AI Strategic Headquarters within the Cabinet Office, chaired by the Prime Minister, which first met in September 2025, approving an AI Basic Plan in December, anchored by a five-year, JPY1 trillion (USD6.3 billion) public investment commitment beginning in fiscal year 2026.
This framework carries no impulsive or looming sanctions, instead imposing only a “co-operation duty” on AI business users to align with emerging government policy.
Elevation Layer
Where the Act sets the direction, voluntary guidelines provide the gradual lifting. For example:
Article 13 of the Act allows the government to establish guidelines ensuring appropriateness in AI research, development, and use. Under that power, the AI Strategic Headquarters published the Guidelines for Ensuring Appropriateness in December 2025.
The Digital Agency’s government procurement guidelines (May 2025) establish a chief AI officer in every ministry and a procurement checklist that evaluates AI suppliers on governance, data handling, output quality, and risk management.
The AI Guidelines for Business version 1.1 (METI and MIC, March 2025) is currently the most detailed articulation of government expectations, organizing guidance around three actor categories – AI developers, providers, and business users – while applying a risk-based approach that scales governance to each system’s capabilities and context.
A separate AI Contract Checklist (METI, February 2025) provides clause-by-clause guidance for structuring AI development and service agreements across three procurement models.
These instruments are not legally binding but are not optional either. Over time, compliance with guidelines may increasingly be treated as evidence of reasonable care under Japan’s general tort framework (Civil Code, Articles 709 and 715).
Enforcement Backstop
In areas where guardrails are well tested, Japan’s existing statutes apply to AI systems with full force. Prominent examples include:
- Act on the Protection of Personal Information (APPI) governs data collection, purpose limitation, cross-border transfers, and prompt inputs;
- Copyright Act governs training data and output infringement;
- Product Liability Act imposes strict liability for AI embedded in products;
- Unfair Competition Prevention Act protects trade secrets;
- Antimonopoly Act addresses competition concerns.
Hard Edges
The perception that Japan is “soft on AI” masks a few sharp enforcement edges. Four areas deserve particular attention:
- Data Protection: The APPI applies where personal information is involved across every stage of the AI lifecycle, with proposed amendments introducing administrative monetary penalties.
- Copyright’s Safe Harbour: The broad exceptions for AI training under the Copyright Act are being tested, leading to lawsuits against AI companies for unauthorized use of copyrighted materials.
- Competition Enforcement: The Japan Fair Trade Commission is scrutinizing market concentration and AI services, particularly in relation to data access and algorithmic collusion.
- Sector Regulators: Various agencies are translating general AI guidelines into sector-specific expectations.
What to Do Now
Six priorities stand out for businesses in Japan’s AI market:
- Map roles under the AI Guidelines and implement governance expectations for each role.
- Build an AI inventory with internal risk classifications.
- Structure vendor contracts using METI’s contract checklist, focusing on input data handling and audit rights.
- Implement trade secret controls for AI tool use.
- If selling to government, prepare for the procurement checklist.
- Engage with evaluation frameworks from the AI Safety Institute for research labs.
Looking Ahead
Several developments will shape the near-term landscape. The APPI amendment bill could introduce Japan’s first administrative monetary penalty regime for data protection violations. The Principle Code for advanced AI may impose more concrete expectations on developers, while ongoing court proceedings will clarify the boundaries of copyright training exceptions.
Japan’s AI governance is not incomplete but deliberately adaptive. It promotes innovation while preserving the capacity to regulate forcefully when needed. The question for businesses is not whether Japan’s approach will harden, but at what pace and in which areas.
