REFERENCE ARCHITECTURE · OPEN SPECIFICATION · RUNTIME GOVERNANCE

AI Governance Control Plane (AGCP)

Deterministic governance at the execution boundary

AI proposes. The control plane evaluates. Authorized execution proceeds.

AGCP is an implementation-independent architectural and specification framework for governing consequential actions in AI-enabled and other programmatic systems. It separates proposal generation from authority to commit an action, establishing deterministic checks at the point where a proposed change may affect an operational system.

The engineering problem

Conventional AI governance measures often concentrate on policy documentation, model behavior, evaluation before deployment, or monitoring after an event. These activities can be useful, but they do not by themselves establish whether a particular consequential action remains authorized and admissible at the moment of execution.

Runtime environments introduce time-of-check/time-of-use gaps, changing authorizations, mutable state, replay risks, and interactions across tenants or governance domains. AGCP addresses these as execution-governance problems rather than relying on a probabilistic model to act as the final authority.

A model’s recommendation is not an execution authorization; a successful tool call is not, by itself, proof of governance conformance or a verified operational outcome.

How the control plane works

The published architectural overview describes a governed action progressing through proposal qualification, a governance decision function, execution eligibility, and governance realization at the commit boundary. Continuation-integrity controls apply to appropriate nonterminal proposals. Governance evidence supports the lifecycle across these stages rather than appearing only at its conclusion.

1. Qualify a proposal

Establish the proposal’s identity, relevant context, and required governance inputs before it is considered for governed execution.

2. Evaluate governance

Apply deterministic eligibility and invariant checks under the controlling specification and adopted governance configuration.

3. Govern commitment

Re-establish execution eligibility at the commit boundary, rather than treating an earlier approval as indefinitely valid.

The precise normative sequence and obligations are defined by AGCP’s controlled requirements, Core Specification, and any expressly adopted normative Companion Specifications. This page is an explanatory overview, not a substitute for those artifacts.

Core design principles

Deterministic lifecycle and ordering

Governance-significant events have ordered, auditable stage records. Lifecycle state is derived from canonical records rather than an independently mutable status label.

Execution-bound authorization

Commit-time evaluation addresses authorization drift and time-of-check/time-of-use risks before consequential execution is permitted.

Structural separation

The control plane’s structural invariants remain distinct from tenant-defined policy logic and from probabilistic proposal generation.

Evidence, isolation, and traceability

Governance evidence, tenant/domain isolation, controlled specifications, and conformance mappings support independent technical scrutiny.

Specification and implementation status

Published specification: AGCP v2.1.0 is identified in the repository as a Public Review Controlled Baseline, dated August 14, 2026. The repository also contains a closed v2.1.x synchronization change set for downstream implementation, but the README expressly states that it is not a separately published release.

Engineering and verification: The program includes a Rust implementation effort, formal test cases, conformance artifacts, standardized test controls, and traceability work. The specification cautions that certain test or injection capabilities may be partially implemented and that any conformance claim must be supported by the applicable profile’s required objective evidence.

This distinction is intentional: an architectural specification, implemented control, completed test, and independently evidenced production outcome are different claims. No blanket certification or production deployment claim is made on this page.

Technical resources

The repository provides the controlled requirements and Core Specification, the Architecture Reference Model, interface contracts, governance-ledger and lifecycle documents, requirements traceability, and conformance materials.

Technical collaboration

I welcome substantive discussions about execution-bound authorization, deterministic runtime governance, assurance methods, architecture review, conformance evidence, and integration with consequential AI-enabled systems.