Here, in a nutshell, is why using AI for military purposes is worse than dangerous:

If you ask it for a plan to attack something, it will never respond "that's bad idea, don't do it."