From Hiroshima,
making AI decisions verifiable.
The Hiroshima AI Assurance Council is an independent, industry-led initiative currently being established in Hiroshima, Japan. It develops HAAP as a proposed practical framework for AI-assisted decisions: verify the required conditions before adoption, then preserve the evidence chain after.
広島から、AI判断の採用条件・証拠・責任の流れを後から確認できる形に残すための、設立準備中の民間・独立の協議会です。From governance principles
to operational assurance.
The Council focuses on the implementation gap: how organizations can prove that an AI-assisted decision met defined conditions before changing real-world state.
ONZALINX, Inc., a Hiroshima-based logistics DX company, leads Council coordination and field implementation.
GhostDrift Mathematical Institute, Inc. initiated the project and contributes Responsibility OS, ADIC, HAAP materials, and formal verification work.
Companies, universities, public organizations, and engineers are invited to contribute through review, PoC testing, and joint research while formal membership terms are being developed.
Assured before action.
Verifiable afterward.
HAAP is a proposed assurance and interoperability layer connecting Business Profiles with agent, tool, and transport protocols. It does not replace HTTP, gRPC, Agent2Agent (A2A), or Model Context Protocol (MCP). It defines how requirements, accountability-relevant information, evidence, verification results, adoption states, and transition authorization records are linked and exchanged.
Required conditions and evidence are evaluated before an AI-assisted decision is adopted as an organizational decision.
Inspection claims are evaluated against evidence independent of the claim, under trust requirements defined by the applicable profile.
Requirements, accountability-relevant information, evidence references, verification results, adoption states, and transition authorization records are integrity-bound and preserved for later review, independent verification, and replay.
From a tested logistics PoC
toward a common assurance protocol.
The first public reference PoC, completed as Phase 1, tested fixed ruleset binding, evidence inspection, independent recalculation, disposition generation, chain-stop control, digest binding, and replay verification for a pharmaceutical cold-chain handoff scenario.
Phase 1 implementation
The public repository documents the implemented functions, test scenarios, replay procedure, scope, and known limitations.
Relationship to HAAP
The PoC provides functional precursors that inform HAAP. It does not implement the complete HAAP record schema, domain-independent state separation, or multi-system interoperability.
Open materials make
the technical basis inspectable.
The Council publishes non-normative technical overviews, formal foundations, and reference implementation materials so that its technical claims can be examined independently. PoC results, implementation scope, and known limitations are documented alongside the relevant projects.
How requirements, accountability-relevant information, evidence, verification results, adoption states, and transition authorization records are linked and exchanged.
Formal Foundations
Lean 4 formalizations of evidence and responsibility preservation under composition, and policy-relevant inspectability.
Working PoCs, replay materials, PoC results, implementation scope, and known limitations.
These materials support independent inspection and reproducibility. They do not by themselves establish the safety, compliance, or correctness of any real-world deployment.
Contribute review,
PoC testing, or domain expertise.
Council inquiries are currently routed through a temporary GhostDrift technical contact form and coordinated with ONZALINX, the representative organization.