Arthur Sidney (The Sidney Group PLLC) has posted Who Can Stop the Machine? Institutional Authority and Meaningful Human Control in AI Governance on SSRN. Here is the abstract:
AI governance often asks whether a human remains “in the loop,” but human presence does not establish meaningful control. This paper argues that meaningful human control is an institutional condition: a responsible person or body must receive relevant information, possess function-specific authority, have practical and technical means to intervene, and retain that capacity as systems, vendors, and dependencies change. The paper develops a six-part Institutional Authority Test covering authority, intervention authority, accountability, runtime governance, coordination, and public legitimacy. It applies the test across procurement, deployment, runtime operation, incident response, agentic delegation, AI-assisted employment decisions, and recent model-containment incidents. The analysis distinguishes technical containment from institutional governance and develops “authority drift” to describe the gradual transfer of practical decision power to AI systems or external providers without an explicit decision to surrender control. It also shows that authority may be distributed or polycentric rather than centralized, while still requiring a traceable path from evidence to decision to enforceable action. The paper concludes with a seven-step operating model for preserving meaningful human governance over time. For sovereign AI and AI governance more broadly, the central question is therefore not merely who owns the model or infrastructure, but who governs each consequential function, with what information, and with what power to intervene.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum