Human Authority Architecture

The Human Authority Architecture is a technical design specification and governance framework for preserving meaningful human authority over autonomous and agentic AI systems. The architecture defines a five-layer structure comprising human authority, AI governance, execution, an enforcement boundary, and infrastructure. Three independent control rails govern authorization, evidence, and stop/revocation. An external witness provides independent verification of system evidence without exercising operational authority. The specification establishes eight architectural invariants and four institutional controls. Its central enforcement principle is that an agent's effective capabilities must remain within the authority granted by the governing policy: E(s) ⊆ A(G(s),s) The design addresses bounded delegation, capability containment, prevention of agent self-amendment, reserved human decisions, revocability, write-ahead evidence, independent verification, and meaningful human control. It also addresses institutional accountability and the risks of transferring consequential judgment to automated systems. The publication includes the integrated technical specification, textual and graphical visual companions, and a technical examination guide with a reviewer response sheet. Release status: Version 0.5, working draft, released for external technical review. The architecture is a design specification, not a certified implementation, a proven operational system, or a statement of legal compliance. The Door Demo exercises only a declared subset and does not establish full architectural conformance. The design has not been independently validated. Deployment-specific engineering, governance, and institutional decisions remain open as identified in the specification.