Admissibility Failure: Why Post-Hoc Governance Cannot Control High-Velocity Socio-Technical Systems

Abstract

Contemporary AI governance frameworks emphasise monitoring, auditing, transparency, and ethical review, yet in practice these mechanisms often generate documentation and visibility without timely control over high-risk behaviour. While these mechanisms improve visibility, they systematically fail to prevent harm in high-velocity socio-technical systems, a pattern repeatedly observed in healthcare safety investigations and financial enforcement actions where risks were known but not acted upon. Drawing on evidence from healthcare safety investigations, financial regulation, public-sector governance, and AI deployment failures, this paper argues that the dominant failure mode is not misalignment, bias, or lack of oversight, but admissibility failure—the absence of explicit pre-execution authority determining what systems are permitted to act, under what conditions, and with what mandatory stop-rights. Post-hoc governance mechanisms—audits, escalation pathways, and remediation—arrive after influence has already occurred, rendering control retrospective rather than preventative. This paper introduces admissibility as a distinct governance primitive, separate from compliance or ethics, and argues that without it, governance mechanisms collapse into documentation rather than decision authority, thereby explaining recurring failures across domains and offering a foundation for designing systems that can halt action before harm propagates.

Ähnliche Arbeiten