Authority Contraction and Refusal as Safety Invariants in Autonomous Systems
Forbes, David
Abstract
This publication is part of a Broader work (full dissertation / SAB 2026):Authority, Refusal, and Resilience in Autonomous Systems — DOI: 10.5281/zenodo.18431599 • Record + PDF: https://zenodo.org/records/18431599 This publication:In this publication, the Stable Authority Boundary is formalized as a design invariant governing safe system behavior. It formalizes the principle that authority in autonomous systems must be explicitly bounded, degrades under uncertainty, and must never escalate implicitly. Legitimate execution of critical actions depends on validated coordination conditions, and refusal to act under insufficient authority is treated as correct and necessary behavior rather than failure. The work introduces and clarifies core concepts including authority contraction, refusal modes, quorum-gated action eligibility, and degraded operational states, without disclosing implementation details, enforcement mechanisms, or control architectures. This publication is intended as a defensive technical disclosure and conceptual reference to establish terminology, define safety boundaries, and serve as prior art for governance-first autonomous system design. It applies across software, cyber-physical, and human-supervised systems. A complete formal treatment, including expanded reasoning and architectural context, is contained in the copyrighted work A Marathon of Restraint: Why Authority Must Decay in Autonomous Systems. Normative companion / testable requirements:Stable Authority Boundary (SAB) v1.0 — Conformance Specification — https://zenodo.org/records/18652047Defines SHALL/SHOULD requirements and conformance assertions for preventing illegitimate execution under degraded coordination. Keywords: Stable Authority Boundary (SAB), authority contraction, refusal as safety invariant, autonomous systems, distributed systems, failure modes, silent failure, system boundaries, authority boundaries, autonomous systems governance, authority limitation, safety invariants, refusal as safety behavior, system legitimacy, governance-first autonomy, execution authority, system authority, AI systems safety, distributed autonomous systems, resilience engineering, fail-safe design, refusal as safety behavior, system boundaries, authority vs control, governance-first autonomy, trustworthy autonomy, autonomous decision-making, failure modes, #EngineeringSeams, #Authority, #DecisionBoundaries, #Refusal, #Silence, #Compliance, #Auditability #DistributedSystems #AutonomousSystems