The Authority Invariant: Why Autonomous Systems Require Explicit Grants of Power

Autonomous AI systems increasingly act in domains where factual commitments and decisions carry real-world consequences. Yet most agentic architectures lack an explicit representation of authority, implicitly equating capability with legitimacy. This foundational paper establishes the Authority Invariant: in autonomous systems, authority must be explicitly granted, bounded, attributable, and revocable prior to action. Responsibility is not an emergent property of system behavior and cannot be reconstructed post hoc from logs or explanations. The paper positions authority as a constitutional prerequisite for trustworthy autonomy and demonstrates why implicit authority leads to hallucinated power, responsibility collapse, and systemic risk. It situates authority as the root article in a broader constitutional framework for autonomous systems, upon which deferral, trust representation, and auditability depend. Implementation mechanisms are intentionally excluded.