I glad you bring up the logistics issues regarding an AI war. It's exactly the same argument I bring up when people in gun control arguments start talking about bringing a gun to a drone fight. IMO the most pressing AI issue isn't run away smart AI, the very real problem now is people falling between the cracks of dumb AI.
We live in very bureaucratic systems; when these were run by humans then bad things happened at the speed of humans and when you detected a bad behaviour a human (manager) could override the regular process. Hopefully. Now, do this at the speed of computers and a negative outcome could be amplified in milliseconds and there is no detection and nobody to apply to. Even worse is that there are a lot of competing AIs made by different actors; what happens when you fall into the cracks between them? It's going to happen.
My gut feeling (not over considered btw) is that we should be doubling down on human rights. If an strange interaction occurs between AIs that results in somebody losing the ability to live a dignified life then it's up to the companies to fix it - at AI speeds - not wait until the office opens on Monday.