The main point when we examine this question should be that the horses already left the stable and we can't stop the advancment of the AI technologies.
Now - I heavily age with this Dmitri - offensive usage of AI will meet defensive counter use and the defensive side can look four vulnerabilities just the same, and fix them.
Criminals are usually less resourceful than countries so we should indeed be less worried about acts and more about the acts of countries- thirds makes me think that the UN should draft a convention for the usage of AI in offensive actress between countries - this is a wild field without rules right now and having internationally accepted set of rules might diminish the extent of the consequences of AI usage.

Another interesting aspect you might investigate in a future article is the defenses needed in order to make sure the technology won't develop a mind of her own and refuse obeying the creator - which can lead her to attack the creator itself in a worst case scenario.

Thanks for an interesting article!