Human authority
Authority should be legitimate, bounded, reviewable and renewable. Capability must not silently widen it.
CONSTITUTIONAL ARTIFICIAL SYSTEMS ARCHITECTURE is PGI’s developing framework for keeping consequential intelligent actions under legitimate human authority.
FRAMEWORK IN DEVELOPMENTCASA separates intelligence from authority. General intelligence does not confer general authority. Superintelligence does not confer sovereignty.
The framework aims to make purpose, permission, human oversight, jurisdiction, audit and remedy part of the action path.

Establish the person, institution, application or agent and the intended purpose.
Determine relevant scope, permission, jurisdiction and current authority.
Keep informed human approval and the exact intended consequence connected.
A protected action should proceed only through the applicable authorized path.
Revocation, expiry, uncertainty and failed verification require appropriate restriction or escalation.
Keep review, explanation, remedy and meaningful non-AI alternatives available.
Authority should be legitimate, bounded, reviewable and renewable. Capability must not silently widen it.
People, institutions and jurisdictions retain their distinct responsibilities. Coordination is not a transfer of sovereignty.
People affected by consequential actions need meaningful ways to challenge outcomes and seek human review.