I don't think an AI has to kill people to be super dangerous. What if a social AI was used by some group to sway/manipulate public opinion on some important iss

I don't think an AI has to kill people to be super dangerous. What if a social AI was used by some group to sway/manipulate public opinion on some important issue? How much manipulation would it take to be dangerous?