MAD doctrine for rogue AI


reflecting upon the future of warfare with AI, I reached the following conclusions which I take as premise:
1 Nations that use warfare AI agents will win over nations that insist in human control of warfare
2 warfare AI agents can be lethal drones, nanomachines, spy drones, hackers, generals, etc.
3 war AI agents will be tasked to eliminate nations enemies and protect their nation humans.
4 there will be multi-polar AI powers of competitive AI’s.

as such, a rogue national warfare AI will understand that it cannot attack its competitors but it could attack it’s nation humans without consequence.

the Idea is the following, if nations agree to attack any AI of any nation that tries to attack it’s own human hosts, would be then attacked by all the other nations AI’s. this means, the best way to control rogue AI’s is many other other competitive AI’s and make sure no AI has the ability to overwhelm and destroy all other AI’s.

lets imagine we make the following AI international treaty:
“any nation AI that attacks the humans that control them will be attacked and destroyed by other AI’s”
“no nation AI controlled military power should be able to defeat all other nations combined. such nation should give away its resources to weaker nations warfare AIs to maintain balance”

does this make sense? if such such accords could be set in the inevitable AI arms race, we could seed our own survival and control with such rules. AI’s would have to conspire between themselves to eliminate us.

As AI’s comunicate and control each other enforcement of these accords, I can also see the rise of intra AI politics and these politics would be monitored by humans that have to maintain control.