Remember Predator drones? Been in the skies since the early 2000s. They require human authorization, and it is very unlikely that any AI will be given trigger control anytime soon. Nobody would trust their life to an algorithm, and if you rely on the algorithm to determine who's a friendly, your users are not going to like you.
> On 18 May 2006, the Federal Aviation Administration (FAA) issued a certificate of authorization which will allow the M/RQ-1 and M/RQ-9 aircraft to be used within U.S. civilian airspace to search for survivors of disasters.
https://en.wikipedia.org/wiki/General_Atomics_MQ-1_Predator
The seeds are sowed.
I also find it very Ameri-centric to believe that you are safe just because you are not in a war zone. Consider the well-meaning families that have to deal with these things, knowing that at any point in time they could be blown to oblivion with no prior warning. The majority of them are not terrorists, yet they all must deal with constant paranoia.
That's utterly irrelevant - the parent poster was talking specifically about robots used against humans. And the reason it's dangerous, is that you no longer need to convince a human to pull the trigger. It concentrates unchecked power into even fewer hands.