As much as I am on the side of Ukraine in their war against Muscovy (Russia), there is one thing that terrifies me; that the drones which are doing most of the killing will become more fully autonomous, and the end result will be armies that can simply commit mass murder by launching the drones in the general direction of the victims, something of a "Skynet" kind of situation, but without the dry one-liners for which Ahnuld is rightly famous.
I would say that we could simply trust that world leaders are sane enough not to do so, but with Vodka attacking mostly schools and apartment buildings, Pooh putting millions of Uyghurs in concentration camps, and Sparkles Khamenei murdering and executing tens of thousands for the "crime" of not liking the current government, I don't think that I can take that assumption to the bank.
So what do we do? Well, the Kyivan Rus has shown us a way to start with massive investments in jamming and anti-aircraft defense, and until Skynet really gets going and becomes more fully autonomous, that will be an effective, if very expensive way of preventing things from going too far. Finding and destroying launch sites will further limit things.
But that noted, I think we have some very real things to deal with with AI, as the Poohs, Vodkas, and Sparkle-Mullahs of the world love the idea of projecting power without needing to persuade people.