Two Structural Boundaries of Governance in Autonomous Systems

This work presents a control-oriented view of AI governance, formalizing the separation between proposal generation and execution authority in autonomous systems. It identifies two structurally distinct governance boundaries: (1) the admissible state space, which constrains the space of proposals that can be generated, and (2) the commitment boundary, where execution authority is deterministically evaluated and enforced. Existing governance mechanisms predominantly operate at the commitment boundary, resulting in reactive control models that filter outcomes without constraining upstream generative capacity. The paper introduces the Principle of Dual-Boundary Governance, asserting that governance is complete only when both boundaries are constrained, ensuring that authority shapes not only what is executed but also what is structurally possible within the system. This framework positions governance as an intrinsic property of system architecture rather than an external layer of policy enforcement.