From Principles to Verifiable AI Governance: Japan's Next Institutional Challenge

Abstract

Around the world, AI governance is moving from high-level principles toward verifiable operational requirements. Japan offers a useful case for observing this transition. While Japan's AI governance has evolved into a multi-layered regulatory stage encompassing the AI Law, the Artificial Intelligence Basic Plan, and government procurement practices, verifiable technical requirements spanning high-risk AI—such as those codified in the EU AI Act—remain fragmented and limited. This article identifies accountability boundaries, safe-state transitions, audit trails, and human oversight as the critical verifiable requirements that must be institutionalized next. Furthermore, it frames the GhostDrift technology suite as a set of proposed implementation candidates to meet these evolving regulatory demands.

1. Japan as a Case: A Multi-Layered but Still Fragmented Governance Structure

Japan's AI regulatory landscape has moved beyond mere principles and guidelines, entering a transitional period where legal frameworks and soft laws coexist. To accurately assess the current state of these regulations, it is necessary to clearly distinguish the following hierarchy of documents:

1. Top-Level Law and National Plans: With the enactment of the Act on the Promotion of Research, Development, and Utilization of Artificial Intelligence-Related Technologies (fully enforced on September 1, 2025) and the Cabinet's adoption of the Artificial Intelligence Basic Plan, Japan's AI policy has entered the stage of developing top-level frameworks that balance technological promotion with appropriate safeguards. However, unlike the EU AI Act, Japan has yet to establish a comprehensive legal regime mandating cross-sectoral technical requirements for high-risk AI.

2. Cross-Ministerial Soft Law: The AI Guidelines for Business (Version 1.1) by METI and MIC emphasize the importance of accountability, human intervention, and verifiability (e.g., considering the recording and preservation of development and inference logs). Although it sets a trajectory for verifiability, transparency, and accountability, it remains fundamentally premised on voluntary compliance.

3. Government Procurement Practices: Documents such as the Digital Agency’s guideline, The Guideline for Japanese Governments’ Procurements and Utilizations of Generative AI for the sake of Evolution and Innovation of Public Administration, detail specific requirements tailored for government information systems, including "mechanisms to withhold harmful outputs," "explainability," and "log acquisition requirements." This tier contains the most specific demands, but it remains centered on the government sector and has not yet evolved into a cross-sectoral regulatory framework.

4. Study Group Documents: Interim summaries from the AI Strategy Council and the AI Institutional Research Group highlight the necessity of legislation to address systemic risks. While they indicate the direction of future regulatory reviews, they remain at the proposal and deliberative stage.

Japan's AI regulatory framework is not a vacuum devoid of ideals; rather, it is in a transitional phase where ideals and practical requirements are distributed across multiple layers. The challenge is not an absence of rules, but rather the need to consolidate these disparate elements into standardized technical requirements for all high-risk AI systems.

2. Comparative Legal Analysis: The EU AI Act's Requirements and the Gap with Japan

Defining the requirements for verifiable governance necessitates a precise comparison with the EU AI Act. Our focus here shifts from broad metaphors to the concrete operational functions mandated by these provisions.

For high-risk AI systems, the EU AI Act mandates the technical capability for automated event recording throughout the system's lifecycle (Article 12), necessitates human oversight (Article 14), requires a quality management system (Article 17), and enforces post-market monitoring (Article 72). Crucially, these are not mere guiding principles; they are binding regulatory requirements inextricably linked to conformity assessments, technical documentation, and operational monitoring.

The Regulatory Gap Between Japan and the EU: Through instruments such as the AI Guidelines for Business, Japan recommends log preservation and accountability, thereby establishing a regulatory "trajectory." In contrast, the EU places automatic log generation and post-market monitoring as legal "requirements." The divergence lies not in the presence or absence of these ideals, but in their degree of institutional codification.

3. Verifiable Requirements Japan Should Institutionalize Next

Drawing on this regulatory gap and the precedents set by government procurement practices, the verifiable requirements Japan must next codify into a cross-sectoral framework fall into four categories:

A. Records and Trails

• Automatic log acquisition

• Integrity of audit trails

• Standardization of technical documentation

B. Operational Safety

• Conditions for output refusal

• Conditions for safe-state transitions

• Continuous monitoring of high-risk systems

C. Human Oversight

• Human-Machine Interface (HMI) standards to operationalize human oversight

• Clarification of intervention and override authority

D. Accountability Structure

• A priori codification of accountability boundaries

• Reporting and re-verification responsibilities during incidents

Moving forward, Japan does not need additional abstract principles. Instead, it must consolidate existing, fragmented practical demands—such as logging, explainability, verifiability, governance structures, and harm mitigation—into unified regulatory requirements applicable to all high-risk AI.

4. Proposed Connection of the GhostDrift Technology Suite to Regulatory Requirements

In translating these ideals into technically verifiable specifications, the components of the GhostDrift architecture can be organized as comparative proposed candidates for these regulatory requirements. Below, we outline how each technology maps to these specific demands.

4.1 Audit Trail Integrity and ADIC

ADIC (Arithmetic Digital Integrity Certificate) serves as a proposed implementation candidate that elevates standard regulatory recording and preservation mandates into the realm of re-verifiability and cryptographic trail integrity. It bridges the gap between the flexible recording methods suggested by Japanese guidelines and the stringent automated logging and regulatory access mandated by EU law, fulfilling the need for "objective, reliable evidence."

4.2 Output Refusal, Passing Conditions, Safe-State Transitions, and UWP

UWP (Unforgeable Watermark Pass) functions as a gating layer that predefines release conditions, refusal triggers, and protocols for transitioning to a safe state. While not explicitly mandated by current legislation, UWP aligns closely with the demands for harm mitigation, operational control, and verifiability inherent in government procurement standards.

4.3 Human Oversight and Beacon

Beacon is best understood not as a standalone Human Oversight mechanism, but as an auxiliary design layer. It empowers human operators to interpret outputs, assess priorities, and make informed intervention decisions. It represents a candidate for Human-Machine Interface (HMI) design, facilitating effective human monitoring and intervention in autonomous AI behaviors.

4.4 Finite Closure as a Design Philosophy to Prevent Ex Post Facto Shifts in Accountability Boundaries

While "Finite Closure" is not an explicitly recognized legal concept under current systems, it serves as a critical design philosophy aimed at preventing the ex post facto shifting of accountability boundaries. Consequently, rather than a technology strictly tailored to existing rules, it is more aptly positioned as a theoretical framework for codifying accountability boundaries—a paradigm future regulatory designs could adopt.

Furthermore, while the Algorithmic Legitimacy Shift (ALS) concept provides a theoretical backdrop for this architectural regulatory shift, this article references it strictly as a supplementary heuristic for understanding regulatory evolution [^1].

5. Conclusion

Through its multi-layered structure of top-level legislation, national plans, soft laws, and procurement guidelines, Japan's AI regulatory framework already champions the ideals of transparency, accountability, verifiability, and human-centricity.

The next imperative for Japan's AI governance is not the proliferation of new principles, but the consolidation of existing, fragmented practical demands into verifiable requirements applicable to all high-risk AI. In this context, the GhostDrift technology suite merits comparative study as a set of proposed implementation candidates for a more verifiable regulatory architecture.

References

• Government of Japan. (Promulgated June 4, 2025). Act on the Promotion of Research, Development, and Utilization of Artificial Intelligence-Related Technologies (Act No. 53 of 2025).

• Cabinet Office. (December 23, 2025). Artificial Intelligence Basic Plan.

• Cabinet Office AI Strategy Council / AI Institutional Research Group. (February 4, 2025). Interim Summary of the AI Institutional Research Group.

• Ministry of Economy, Trade and Industry (METI) & Ministry of Internal Affairs and Communications (MIC). (March 28, 2025). AI Guidelines for Business (Version 1.1).

• Digital Agency. (May 27, 2025). The Guideline for Japanese Governments’ Procurements and Utilizations of Generative AI for the sake of Evolution and Innovation of Public Administration (Digital Society Promotion Standard Guidelines DS-920).

• European Parliament and Council. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence.

Supplementary Materials

• Manny. (2025, 2026). Mathematical Modeling of the Ghost Drift Phenomenon and Theoretical Examination of ALS (Algorithmic Legitimacy Shift). (Unpublished manuscript).

[^1]: ALS (Algorithmic Legitimacy Shift) is a theoretical model positing a phase transition where the regulatory center of gravity shifts from principle-centric justification to passing-condition-centric justification. Under the condition of $B < J$ (where $B$ is the system baseline and $J$ is the justification threshold), legitimacy is transferred into an irreversible regime. In this article, it is not used for the direct interpretation of regulatory texts, but referenced solely as a supplementary heuristic to observe regulatory paradigm shifts.