Institutional Friction Toolkit

Auto-Loop Invalidity Rule

Review circular self-approval by automated systems.

This rule applies to AI systems involved in iterative self-improvement, automated testing, or performance optimization pipelines.

It does not prohibit development. It defines conditions under which approval is procedurally invalid.

External Explanation

Closed-loop self-assessment occurs when the same AI system, or a dependent automated loop, generates criteria, evaluates outputs, and approves continuation without independent human verification.

For LUMINA-30 boundary review, this is insufficient evidence of effective human refusal because the system is materially assessing its own continuation path.

Closed-loop self-evaluation means that the same AI system, or a dependent automated loop, generates the criteria, evaluates the output, and approves continuation without independent human validation.

Under LUMINA-30 boundary review, this is not enough to demonstrate effective human refusal, because the system is effectively reviewing its own continuation path.


1. AI Evaluation Non-Sufficiency

AI-generated evaluation results, performance metrics, or optimization outputs shall not constitute sufficient grounds for approval.

Human written justification is mandatory.


2. Human Judgment Requirement

For each major release, update, or deployment:

If no written reasoning exists, approval is invalid.


3. Closed-Loop Prohibition (Procedural)

If an AI system:

without independent human validation,

the approval shall be considered procedurally invalid.


4. Reversibility Confirmation

Before deployment, confirm:

If reversibility cannot be demonstrated, approval is invalid.