Beyond Policy-as-Code: Admissibility as a Runtime Primitive for Autonomous Systems

Autonomous AI systems increasingly generate their own execution paths, selecting tools, composing actions and proposing state changes dynamically. This technical paper examines the runtime governance problem created by that shift. Constitutional Computing does not replace established security mechanisms such as least privilege, capability security, reference monitors, Zero Trust or policy-as-code. It builds on those foundations and examines how their enforcement requirements apply when autonomous systems generate consequential execution paths dynamically. The paper develops the distinction between capability, permission, authorization and admissibility, and argues for deterministic consequence boundaries without requiring deterministic reasoning. Domain: Execution governance for autonomous AIObjective: Governed Autonomy Published by KeyWow.