Well, that's all very idealistic, but it's likely not going to happen.

Israel already used AI to pick bombing sites, those bombs and missiles would have been programmed with altitudes and destinations (armed) then dropped. The pilots only job these days is to avoid interception, fly over the bombing locations, tag the target when acquired, and drop them. Most of this is already done in software.

Eventually humans will leave the loop because unlike self-driving cars, these technologies won't risk the lives of the aggressor's citizens.

If the technology is seen as unstoppable enough, there may be calls for warnings to be given, but I suspect that's all the mercy that will be shown...

... especially if it's a case of a country with automated technologies killing one without or with stochastically meaningless defenses (eg. Defenses that modelling and simulations show won't be able to prevent such attacks).

No, in all likelihood the US will tell the country the attack sites, the country either will or will not have the technical level to prevent an amount of damage, will evacuate all necessary personal, and whoever doesn't get the message or get out in time will be automatically killed.

Where defenses are partially successful, that information will go into the training data for the next model, or upgrade, and the war machine will roll on.