A kill switch may be effective for an agent driven attack or something. But I would think that what ever hypothetical national security endangering threat that was created would be something a deterministic program would handle, not an agent driven thing. So implementing a kill switch would have no effect on the hypothetical deterministic program that was created.