AI Guidelines (v1.1) Compliance Mapping | GhostDrift

GhostDrift Research Institute has published its current compliance mapping for the Japanese Government’s AI Guidelines (Version 1.1).

This is not a declaration of full compliance.It is a structured disclosure of what has been technically implemented, what evidence is publicly verifiable, and which organizational layers remain under development.

Our Position

In AI governance, the critical distinction is between:

• “We explained it.”

• “It can be independently verified.”

GhostDrift takes a clear position:

Accountability must be grounded in recomputability, not explainability.

Under our Responsibility Engineering framework, we structurally fix the locus of responsibility through:

• Pre-defined decision boundaries

• Automatically generated Fixed Certificates

• Append-only, tamper-evident ledgers

This architecture ensures that decisions are not merely described but can be deterministically recomputed.If recomputation conditions are not satisfied, the system triggers ABORT (Stop Boundary).

What Is Already Implemented

The following guideline items are technically implemented at the architectural layer:

• C-6.1 (Transparency)

• C-7.1 (Traceability)

• C-7.6 (Documentation — partial implementation)

Specifically, the public implementation generates:

• certificate.json (fixed execution conditions and verdict)

• ledger.csv (append-only verifiable ledger)

• evidence_timeseries.csv (deterministic recomputation evidence)

These artifacts allow third parties to reproduce the same OK/NG result from the same input conditions.

Public repository:https://github.com/GhostDriftTheory/ghostdrift-adic-audit

What Is Still Under Development

Organizational layers are currently being structured, including:

• Formal privacy policy implementation

• Security SLA and vulnerability disclosure procedures

• Red-teaming evaluation pipeline

• Bias assessment process

These items are explicitly marked as “in progress” in our mapping page.

Why We Publish This

Compliance should not rely on declarative statements alone.

By separating:

• Mathematical verification layers

• Organizational policy layers

we provide a transparent representation of our current implementation status.

This publication is not a completion claim.It is a structural disclosure of the present state.

https://www.ghostdriftresearch.com/ai-guideline-mapping