A security separation that prevents an agent from modifying its own control logic or permissions, keeping reflexive changes safe and auditable.