ASiSO-I – Action-Specific Oversight for Industrial AI Systems A Proportional Governance Architecture for Automation, Safety, and AI-Driven Operations under the EU AI Act with ASOSE-Based Semantic Readiness as a Legitimacy Condition

Industrial AI systems increasingly operate across tightly coupled action chains that link perception, prediction, optimization, and physical or organizational intervention. Within the same industrial environment, AI may perform low-risk analytical functions, medium-risk operational recommendations, and high-risk actions affecting safety, continuity of operations, regulatory compliance, and environmental impact. System-level AI regulation is therefore structurally incapable of reflecting the internal heterogeneity of industrial AI usage.

This paper introduces ASiSO-I, an industrial application of Action-Specific Oversight (ASiSO) that binds oversight obligations to specific AI-mediated actions in context, rather than to model identity or system category. Oversight is operationalized through a three-layer architecture (Free Execution, Conditional Oversight, Mandatory Oversight), enabling proportional human intervention only at those action points where industrial risk materializes.

Crucially, the paper demonstrates that action-based industrial governance is legitimate only when grounded in Semantic Readiness, provided by ASOSE (Akarkach School of Semantic Emergence). By integrating ASiSO with semantic readiness, ASiSO-I establishes a gold-standard reference architecture for auditable, safety-aware, and innovation-compatible AI governance in industrial and critical-infrastructure environments under the EU AI Act.