Search for a command to run...
Autonomous systems, including AI agents and automated infrastructure components, are increasingly capable of proposing and executing operational mutations across enterprise environments. Existing governance mechanisms—such as identity and access management systems, policy engines, and Zero Trust architectures—evaluate permissions and policy compliance but do not define a unified execution-layer model governing how and when system state transitions are committed. This paper introduces the Artificial Intelligence Governance Control Plane (AGCP), a deterministic control-plane architecture for governing execution at the point of state mutation. AGCP defines a five-stage governance pipeline, an artifact-centric interaction model, explicit commit semantics, and a ledger-derived lifecycle model in which system state is reconstructed from append-only history rather than maintained as mutable state. The architecture enforces a strict separation between reasoning, governance, and execution, ensuring that all state transitions are mediated by authorization artifacts and lifecycle constraints. Deterministic evaluation and replayability provide reproducibility and auditability, while execution gating ensures that only admissible transitions are committed. AGCP is grounded in distributed systems principles including deterministic state machines, commit semantics, and log-derived state reconstruction. Although motivated by the emergence of autonomous agents, AGCP is not limited to AI-based systems. The model applies to any system capable of proposing or executing state transitions, including traditional software systems, infrastructure automation, and distributed services. AGCP is positioned relative to identity systems, policy engines, Zero Trust architectures, and workflow systems, and is presented as a general execution-layer governance framework for deterministic and verifiable control of system state transitions. AGCP is formulated as a deterministic state-machine control plane in which system state is derived from an ordered log of transitions.