The Custody of Intelligence: Power, Sovereignty, and Human Agency Beyond the Open–Closed Divide

Artificial intelligence has outgrown the debate that dominates its governance. Public and policy attention has organized around whether the most capable models should be released openly or held closed — a real question that no longer reaches the one that matters. As machine intelligence becomes abundant, autonomous, physically embodied, geographically distributed, and capable of directing consequential resources, the defining governance question is not how intelligence is distributed, but who holds custody over it, what authority may be delegated to it, and whether that authority remains accountable to humanity.

The monograph argues that intelligence — civilization's scarcest strategic resource — is becoming abundant, and that this transformation reorganizes economic power. Just as the Industrial Revolution transformed the economics of physical labor, the Intelligence Revolution is transforming the economics of cognition. As the unit price of cognition collapses, its value does not disperse but migrates to what remains scarce: the compute, capital, and energy that produce it. Cheap intelligence does not democratize power — it concentrates it at the physical substrate, where much of the price is scarcity rent. Who captures that value, and whether that concentration is accountable, is the custody question in economic form.

The monograph then shows how authority over intelligence comes apart from the humans meant to hold it along five axes: as its economics concentrates, as it disperses beyond jurisdiction, rises beyond the reach of enforcing bodies, travels beyond the real-time human loop, and acquires an agency that dissolves the assumption that it is merely a tool. These are shown to be one problem — the custody problem — with a permanent human line at its center.

To govern that problem, the monograph proposes the Intelligence Stewardship Framework: a complement to existing capability-focused instruments (the NIST AI Risk Management Framework, the EU AI Act, and frontier scaling policies), occupying the layer of authority and accountability they leave open. The framework classifies systems across three layers held to different standards of evidence, tiers the authority a system may exercise, fixes Prohibited Authority Thresholds that must remain human regardless of capability, and reduces those principles to enforceable constraints — including the result that authority over consequential resources is insurable, and therefore permissible, only when it is bounded.

The monograph does not predict a single future; it identifies the governance questions that remain important across many plausible futures. Its central claim is that the defining question of the coming century is not how intelligent our machines become, but whether we retain meaningful custody over the authority we delegate to them.