the moment some dumb fucker (probably in the US military) leaves a kill-chain-connected AI exposed in any way to the internet, someone else (probably in thigh highs) will find it, break in, and prompt engineer it to do the dumbest possible thing at that particular moment. And as much as I want to believe it would be prompted with something like "kill all the billionaires", they would probably end up starting with something like "gee, I wonder if I could drone strike my neighbor?", get kicked out/patched immediately after, and the perpetrator arrested in short order.
in other words, probably, someday, yes; but actually no.