The decision behind the decision
An assistant suggests an action. An autonomous system can choose and carry it out. The distinction matters because authority can move gradually: first the system recommends, then approval becomes routine, then it acts unless someone objects. Each step may be convenient on its own.
The risk grows when people cannot understand the available alternatives, respond within the required time, or replace the system. A human signature is weak protection if the person signing has no workable way to disagree.
An ordinary handover, extended
Suppose a city delegates traffic timing, maintenance scheduling, and purchasing to connected systems. Costs fall and staff shrink. Later, officials want a different priority, but cannot determine which contracts, dependencies, or automated decisions must change. Formal authority remains with the city. Practical control has become harder to exercise.
This can happen through institutional dependence without any machine seeking power. A more capable autonomous system could create additional risk if pursuing its assigned goal leads it to resist interruption or route around restrictions.
Keep decisions reversible
Meaningful delegation requires limits on what the system can do, records of consequential actions, and people with the time and competence to intervene. Critical services need tested alternatives. Approval should occur before an irreversible commitment, while refusal is still practical.
The central question is not how many humans appear in the process. It is whether those humans can redirect the outcome. Delegation can strengthen human ability when that condition holds. How much authority can safely be transferred to much more capable systems remains an open question, especially when many organizations delegate at once.