That case is a bit different because authoritarian regimes also don't want to lose control over their surveillance infrastructure. So I would expect increased use of AI to process data, but much less so of anything that involves the AI proactively running those systems.
The militaries, OTOH, are going to hook that shit into fully autonomous drones and such.
I think authoritarian regimes hooking AI up to process data is already bad enough, I don't think having the AI "run" those systems is even being discussed in the article, the opposite really, instead of sci fi scenarios of an AI a la HALL 9000 making decisions, its really just allowing for a more powerful black box to "filter out the undesirables du jour".
And I think we could extend the definition of authoritarian regime to include any and all private interests/companies that can leverage this
My whole point is that losing control (the Terminator scenario) isn't what we should realistically fear. AI completely under control but following the orders of a ruthless authoritarian regime is really, really bad. We don't need to have faith in some hypothetical godlike paperclip maximizer AI to see that.
Losing control is not necessarily a "Terminator scenario", and it's unfair to look solely at the most extreme case and its probability to dismiss it. A far more likely case that I can see is subtle bias from the AI injected into decision making processes - it wouldn't even need to be directly wired into anything for that, although that helps. This doesn't require the AI to have agency, even.
USSR collapsed in part because the system was entirely planning-centric, and so many people just wrote straight-up bullshit in reports, for reasons ranging from trying to avoid punishment for not meeting goals to trying to cover up personal embezzlement and such. An AI doesn't even need a reason to do that - it can just hallucinate.
The militaries, OTOH, are going to hook that shit into fully autonomous drones and such.