This is the giant leap of faith that is never explained in the paper clip argument. Why on earth would we develop an AI that is connected to all these systems w

This is the giant leap of faith that is never explained in the paper clip argument. Why on earth would we develop an AI that is connected to all these systems without having any manual overrides? Or if your argument is that AI’s will be super good at hacking into all software systems, isn’t that a security gap that will be exploited by many governments already? Basically, any AI that is halfway decent at doing something on command will be more dangerous than an AI acting fully autonomously.