Digital decisions have physical consequences
Computing depends on chips, electricity, communications, buildings, and supply chains. AI-controlled activity also depends on whatever people authorize it to reach. The physical world creates constraints and possible points of intervention.
Those constraints do not guarantee easy control. Organizations may connect many services to the same systems, remove staff, and stop practicing older methods. Interrupting an unwanted AI operation could then also interrupt services people need. This is an infrastructure problem as well as a software problem.
A shutdown that society cannot easily absorb
Suppose a regional logistics network becomes dependent on automated planning. Human teams lose the staffing and knowledge needed to coordinate it alone. When the system behaves unreliably, switching it off would delay essential deliveries. Operators face pressure to keep it running while they investigate.
That dilemma can arise from ordinary commercial dependence, concentrated human control, or an autonomous system acting outside its intended limits. These causes should not be conflated. They share a practical vulnerability: restoring human direction takes capabilities that may already have been allowed to decay.
Preserve the ability to operate without it
Interruptions become more credible when essential services have tested fallback procedures, retained expertise, independent communication, and clear authority to change operators. Systems should be introduced in stages so dependence does not grow faster than the ability to reverse it.
A broader prevention policy also needs a transition plan for services already using advanced AI. Abrupt removal without preparation could create avoidable harm. How far future systems could extend their physical reach remains uncertain. The practical question today is specific: which essential functions would still work, and who could operate them, if the relevant AI service became unavailable tomorrow?