Not so much "hacked" as "written from the ground up by people with nobody's interests but their own at heart".
AI isn't naturally occurring. Humans write the code, humans select the training strategies, humans choose the data to train it with and most importantly humans choose what constitutes a "good, well-trained" AI over a "bad" one which needs more training. Humans pick when to stop. In this hypothetical, those humans are now the gatekeepers on all governance. They have no motivation to do anything but serve their own bias. And we know how much ethical training programmers receive.
We can't build a perfect AI because we don't have a perfect AI which could build it for us.
AI isn't naturally occurring. Humans write the code, humans select the training strategies, humans choose the data to train it with and most importantly humans choose what constitutes a "good, well-trained" AI over a "bad" one which needs more training. Humans pick when to stop. In this hypothetical, those humans are now the gatekeepers on all governance. They have no motivation to do anything but serve their own bias. And we know how much ethical training programmers receive.
We can't build a perfect AI because we don't have a perfect AI which could build it for us.