From Participation in the Hiroshima AI Process to Contributions Through Technology — The International Role AI Assurance Aims to Play

Turning AI governance principles into independently verifiable technologies. Toward a society where AI-driven decisions can be confidently adopted and executed across national and organizational boundaries.

Moving Beyond Trust Alone in AI Safety

Artificial intelligence is transforming decision-making across businesses, governments, and society.

However, an AI system’s ability to generate sophisticated recommendations does not necessarily mean those recommendations can be safely adopted and executed in the real world.

For example, even if AI proposes an optimal logistics plan, that plan cannot be executed unless the necessary quality and safety requirements are satisfied. Similarly, an AI-generated response in public administration should not be provided to citizens without verifying its supporting evidence and conditions of use.

Improving AI performance and verifying whether AI-generated decisions are appropriate for adoption and execution are two distinct challenges.

We believe that connecting these two dimensions through verifiable technologies will become essential to the future of AI.

Turning the Principles of the Hiroshima AI Process into Practical Technologies

Launched following the G7 Hiroshima Summit in 2023, the Hiroshima AI Process has established international guiding principles and a code of conduct aimed at advancing safe, secure, and trustworthy AI.

The Hiroshima AI Process Friends Group Action Plan 2026, announced in March 2026, further emphasized the importance of advancing AI governance and its practical implementation throughout society.

The Hiroshima AI Assurance Council (HAAC), which AI Assurance, Inc. helped establish and supports as a technical contributor, has joined the Hiroshima AI Process Friends Group’s Partners’ Community.

Hiroshima-based ONZALINX serves as the council’s representative organization.

Within this international framework, AI Assurance, Inc. aims to make a particular contribution:

Connecting the principles of AI governance with technologies that enable those principles to be verified and applied in real-world operations.

Three Roles We Aim to Fulfill

1. Translating AI Governance Principles into Verifiable Operational Requirements

“Ensuring safety,” “establishing accountability,” and “maintaining appropriate human oversight” are fundamental principles of responsible AI.

However, these principles alone do not determine whether a particular AI-generated decision should be adopted in an actual business or administrative process.

We aim to develop technologies that translate these principles into concrete operational requirements, supporting evidence, and clearly defined responsibilities.

Our objective is to make the conditions under which AI decisions may be adopted and executed explicit and verifiable.

2. Enabling Independent Verification Across Different Regulatory Frameworks

Different countries and regions naturally establish different AI regulations, evaluation criteria, and governance frameworks.

The goal should not necessarily be to impose a single universal set of evaluation criteria.

Instead, we believe the international community can benefit from a common technical approach to verification.

Even when countries and organizations apply different AI safety standards, the validity of agreed conditions and propositions can be independently examined through a common verification methodology.

At AI Assurance, we are pursuing this approach through formal verification technologies, including Lean 4.

Formal verification makes it possible to mathematically examine whether explicitly defined properties hold under specified assumptions, with independently checkable proofs.

This does not mean proving that an entire AI system is universally safe or automatically compliant with every applicable law.

Rather, it enables a clear distinction between what has been verified, under which conditions, and what remains unverified.

We believe this distinction is essential to establishing meaningful and internationally interoperable AI assurance.

3. Connecting Verification Technologies to Industry and Public Administration

Even the most sophisticated AI governance framework has limited value unless it can be applied in real operational environments.

AI Assurance, Inc. is developing its Responsibility OS (ADIC) and related technologies to verify the conditions and supporting evidence required before AI-generated decisions can be adopted and executed.

Our publicly available proof-of-concept implementations cover areas including:

• Pharmaceutical cold-chain handover decisions

• Adoption of AI-generated responses in public administration

• Privacy assurance for AI-assisted information processing

• Lifecycle management and usage restrictions for AI-generated information

These initiatives demonstrate how specific assurance requirements can be translated into technical verification processes within defined experimental environments.

Looking ahead, we aim to extend these technologies across logistics, manufacturing, public administration, finance, cybersecurity, and other fields.

Our broader objective is to develop verification infrastructure that can operate across industries and organizational boundaries without requiring every participant to adopt identical AI systems or operational frameworks.

Contributing to AI Safety Through Cyber Assurance Technologies

AI safety cannot be achieved solely by evaluating the behavior of AI models.

As AI becomes increasingly integrated into critical infrastructure and operational systems, it is equally important to examine whether the systems executing AI-driven decisions can operate safely and reliably under changing conditions.

Cyberattacks, system failures, unauthorized operations, and emergency failover scenarios can all affect the safety of AI-enabled operations.

AI Assurance, Inc. is developing Cyber Assurance technologies designed to verify the conditions and supporting evidence required for critical system operations, including failover and other consequential actions.

Our approach aims to verify specified execution requirements and, when integrated into an execution control mechanism, withhold authorization if those requirements are not satisfied.

AI safety must address not only the decisions generated by AI, but also the security, reliability, and execution conditions of the systems that act upon those decisions.

Through Cyber Assurance, we aim to contribute to AI safety from both perspectives: the verifiability of AI-driven decisions and the assurance of the operational systems responsible for executing them.

We believe that connecting AI governance, formal verification, and cybersecurity assurance can help strengthen the foundations of safe, secure, and trustworthy AI.

From Hiroshima to the World — Our Vision

We view participation in the Hiroshima AI Process Partners’ Community not simply as an expression of support for its principles, but as an opportunity to contribute concrete technologies and practical implementation results to the international community.

Connecting international governance discussions with technologies used in real operational environments.

Presenting implementation results in ways that allow independent verification.

And making the resulting knowledge and technologies applicable across national and industrial boundaries.

While the Hiroshima AI Assurance Council provides a collaborative platform connecting diverse organizations, AI Assurance, Inc. aims to contribute as an independent technology developer supporting the practical realization of these objectives.

From a society that uses AI based on trust alone to one that can confidently use AI after verifying the conditions under which its decisions may be adopted and executed.

From the principles of the Hiroshima AI Process to technologies that support real-world decision-making.

Our vision is to make Japan a society where AI can be used with greater confidence than anywhere else in the world — and to share these practices with the international community.

AI Assurance, Inc. will continue advancing research, development, and practical applications toward this vision.

Related Information and Public Technologies

Hiroshima AI Process Friends Group — Partners’ Community (Ministry of Internal Affairs and Communications, Japan)https://www.soumu.go.jp/hiroshimaaiprocess/supporters.html

Hiroshima AI Assurance Council (HAAC)https://hiroshima-ai-assurance.com/

Hiroshima AI Process Assurance Mappingshttps://github.com/GhostDriftTheory/hiroshima-ai-process-assurance-mappings

Participation in the Hiroshima AI Process Friends Group’s Partners’ Community does not constitute certification, accreditation, or endorsement of AI Assurance, Inc. or its technologies by the G7, the Government of Japan, the OECD, or any other participating organization.