Bounded Autonomy

This doctrine formalizes the non-negotiable rules governing execution authority within autonomous systems. These rules hold regardless of intelligence capability, deployment context, or system scale.

Core Doctrine

Intelligence may decide. Execution must be explicitly authorized. Intelligence may reason, plan, simulate, optimize, generate code, and persuade without restriction. It has zero authority to commit irreversible action without authorization.

Structural Separation of Cognition and Authority

Execution authority exists within the system boundary but outside the cognitive domain. Cognition cannot grant, infer, or escalate its own authority. Influence does not equal authority.

Irreversible Execution Boundary

All execution is irreversible in the sense that it cannot be un-executed. Irreversibility is the decisive boundary trigger that determines when authority must apply. No execution occurs without crossing this boundary.

The Ten Laws of Authorized Execution

Execution Envelopes

Authorization determines whether execution may occur. Execution envelopes define the permitted scope, resources, and blast radius of authorized execution. Envelopes cannot be selected, expanded, or modified by intelligence.

Authority Over Accountability

AICE does not rely on accountability to correct behavior after execution. Authority prevents unauthorized execution from occurring at all. Records exist solely to verify that authority rules were obeyed.

Back-Door Elimination

Back doors are not a code problem; they are an authority problem. If an intelligence can both author the system and grant itself execution, back doors are inevitable. Bounded autonomy prevents this class of failure structurally.

Positioning

AICE is not an AI model, agent, application, or alignment system. It is authority infrastructure for cognitive execution.

Canonical Doctrine

The authoritative statement of this doctrine is published as a formal PDF. In the event of any conflict or ambiguity, the PDF is canonical.