Open standard · draft-howlett-aigcsep-00

The AI Governance Control & Service Exchange Protocol.

A proposed open standard for the secure, interoperable, and verifiable deployment of autonomous AI agents. Submitted to the IETF on 2026-06-07. The full text, revisions and discussion live at covenant.to.

The 3-plane architecture

Power in one plane. Governance in another it cannot reach.

The core of the RFC is a strict separation of concerns. The AI operates powerfully in the service plane, but its governance is cryptographically enforced from an external, unreachable governance plane.

Diagram of the AIGCSEP 3-plane architecture showing separation of service, control and governance planes
1 · Service plane

Where the AI model lives and works.

Highly dynamic and performant. The agent does its job here, with no access to the keys that govern it.

Focus: performance and utility
2 · Control plane

Configuration, deployment, policy.

Manages configuration, deployment, and policy distribution between the other two planes.

Focus: configuration and state
3 · Governance plane

The immutable oversight layer.

Logs actions and enforces constraints cryptographically, outside the AI's reach.

AI entityPublic certificate
//
Governance planePrivate key
Focus: the cryptographic boundary

This is what FireControl implements.

Zones, roles, a hardware-key gate for root, a signed stop and an append-only ledger, with a live AIGCSEP compliance level.

How FireControl works