The Governance Decision and Enforcement responsibilities separated in the preceding section are connected to actual Execution Control through the Execution Boundary.

The basic structure is clear.

AI decides. OS / Kernel enforces.

Within the upstream AI / Agent System, Decisions such as Allow, Hold, and Deny are formed based on the Governance State, including Policy, Authority, Context, and other relevant elements. The OS / Kernel below the Execution Boundary does not reassess the reasons behind those Decisions. It receives the upstream Decision and execution conditions and performs Deterministic Enforcement.

As a basic principle, the same Decision under the same execution conditions should therefore produce the same Enforcement Result.

One reason to consider the OS / Kernel Level is its proximity to actual Execution. Actions through which AI affects System Resources or the external world ultimately pass through concrete Execution Points such as Process Execution, File Access, Network Access, Privilege Control, and Device Access.

Even if an AI decides to read a file, transmit data over a network, launch a Process, or operate a Device, the Action must pass through these lower-level System functions before it can actually occur.

Positioning the Execution Boundary at this level makes it possible to enforce the Decision close to the point of actual Execution, rather than merely forming the Decision within the Application. This differs from a structure that relies solely on the AI itself to comply with the Governance Decision.

In the case of Hold, the OS / Kernel does not perform additional inference. Execution remains suspended until the required Governance conditions are resolved. Nor does the OS independently convert an unresolved state into Allow or Deny.

This design keeps Governance Decision Responsibility upstream and Enforcement Responsibility downstream. If a problem occurs, it remains possible to distinguish between “what was decided” and “whether that Decision was correctly enforced.”

The OS / Kernel is not a second Governance brain.

The OS does not decide. It enforces what has already been decided.

This is the fundamental principle of role separation at the Execution Boundary.