I was similarly appalled, but as I read it (WaPo) it was more a matter of letting the AI identify 'targets'; the humans did the rest. They didn't go full-on SkyNet - but they did take the humans out of the loop when identifying targets.

I've read nothing about 'permission to kill 20 civilians', but acceptable 'collateral damage' (what they call this) is part of their rules of engagement. I have zero military background, BTW.