A4-Gov: A Governance Architecture for AI Service Systems

초록

Public AI services often undergo ex ante review before deployment, but the resulting risk decision must still be carried into operation through concrete controls, review triggers, and evidence records. This paper defines A4-Gov as a runtime governance architecture for AI service systems that performs that operational role. The architecture assumes that review or risk grading has already determined the control level required for the service or case. A 4-Gov then carries that decision into four runtime functions: Awareness, which derives the active baseline from policy and service context; Assessment, which identifies governance-relevant deviation from that baseline; Adaptability, which selects bounded intervention and triggers human review when required; and Accountability, which preserves policy state, intervention rationale, and audit-ready evidence. The paper positions the architecture against AI governance frameworks and self-adaptive software architectures, then examines it through a public-service walk-through and a failure-oriented analytical check. The contribution is architectural: it specifies how review decisions, bounded intervention, and accountability logging remain connected during operation.

키워드