Do you want them to wait until an agent gets compromised or goes on a sidequest to test if a determined AI can break containment?