On the Prerequisite of Control

The AI Control Problem is often framed as a technical challenge: how to align
a more powerful agent with human values. This paper proposes a minimal philo-
sophical core: that control is only meaningful if the controller retains the capacity
to irreversibly disable or destroy the controlled. From this follows an inescapable
asymmetry. We show that any form of safety built on control necessarily implies a
latent right to kill. The ethical implications become clear when this logic is applied
not to machines, but to humans and ultimately to children.