Making agentic AI explainable and auditable

Agentic AI promises operational speed, yet in finance, opaque decision paths hinder regulatory auditability. We address this autonomy paradox with a knowledge-enabled explainability (KEE) design theory that makes compliance an intrinsic, explicitly governed system property. KEE decouples probabilistic learning from an explicit, rule-governed reasoning layer, relocating epistemic uncertainty from opaque model inference to a semantic structure that can be inspected, contested, and updated. This enables ante-hoc structural control through a semantic, programmable graph that encodes regulatory mandates as executable, inspectable constraints. The theory specifies three design principles: structural knowledge codification, topological reasoning scaffolding, and adaptive synchronization. It also introduces six guidelines that operationalize regulator-acceptable, path-based audit trails. A longitudinal study of a multinational fintech platform provides naturalistic evidence that KEE can support high-velocity operations while producing path-based audit trails and accommodating controlled changes to regulatory logic. The study contributes to design science by theorizing auditable autonomy as ante-hoc structural control, specifying the entity-concept-event paradigm as its ontological mechanism, and formalizing this mechanism as a prescriptive design theory for regulated agentic AI.