The Five Gates: Why Every Autonomous Agent Action Must Pass ARCEB Before Execution

Most AI governance frameworks describe what they want AI to do. ARCEB describes what AI must prove before it does anything.

ARCEB stands for Auditable, Reversible, Compliant, Explainable, and Bounded. On every 25 Alpha-licensed platform, every autonomous agent action must pass all five gates before execution. Not four of five. All five. No override. No exception.

The Auditable gate requires that the action creates an immutable, timestamped audit record before it executes — not after. The record cannot be modified or deleted by any principal, including 25 Alpha LLC itself. This is a technical constraint enforced at the infrastructure layer, not a policy aspiration.

The Reversible gate requires a pre-execution snapshot or rollback checkpoint for any action that modifies persistent state. Business-critical actions — financial commitments, data deletions, communications — require explicit principal confirmation before the snapshot closes.

The Compliant gate runs a real-time policy check against the platform's active compliance posture. If the action would violate a current regulatory constraint, it stops here.

The Explainable gate requires that the reasoning chain behind the action be serialized and stored alongside the audit record. A human reviewer must be able to reconstruct why the agent took the action without executing it again.

The Bounded gate enforces scope. Every agent operates within a defined authority boundary. An agent authorized to read a document cannot write to it. An agent authorized to draft a response cannot send it. Boundaries are set at deployment and cannot be expanded at runtime.

The ARCEB score published in the 25 Alpha Trust Center is not a self-assessment — it is the aggregate of automated gate pass rates across the platform's active agent population. A score of 94 means 94% of agent actions cleared all five gates on the first pass. The 6% that did not were escalated to human review before execution.

This is what governed AI looks like at the execution layer. Not a policy document. A gate.