Most of the worst things done in history are being done by people who are absolutely convinced that they are doing something VERY good. For evils beyond an individual thief or killer they were almost always committed by someone who believed they were being good.

The lesson from history is that doing good is the most reliable path to do evil. To do actual good: leave other people alone, unless they ask for your help.

"AI safety" practitioners fit the mold of historical great-evil-doers well. They are the true danger in AI.