Authority Before Autonomy: Independent Enforcement Boundaries for Self-Improving and Agentic AI

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Artificial intelligence systems are increasingly capable of autonomously designing architectures, executing experiments, modifying software and participating in recursive self-improvement. This technical research note examines the engineering implications of that progression and asks a distinct question: as autonomous capability increases, what independently determines whether an AI-generated decision is authorised to become an executable consequence? Drawing on research in autonomous architecture discovery, recursive self-improvement, agent governance and pre-execution control boundaries, the paper develops the principle Authority Before Autonomy. It distinguishes access and computational capability from executable authority and examines the need for independently enforceable authority boundaries capable of remaining effective across changing models, distributed systems, degraded connectivity and consequential execution paths. The paper also positions the IrisKey.ai governance architecture within this emerging research landscape, while distinguishing its proposed architecture from earlier software-based authorisation and agent-governance approaches. Keywords: AI governance; agentic AI; recursive self-improvement; autonomous systems; execution authority; AI safety; runtime governance; human authority; independent enforcement; consequential AI. Technical Research Note / Preprint, Version 1.0.