The biggest problem I see with current state of AI is the feedback loop it creates in people.

For example, some (usually older people in higher positions) perceive long emails as a sign of courtesy, involvement, and hard work. As a result, others have started using AI to write longer emails for them, to get on "the good side" with their higher ups. And now that many use AI to generate longer emails, those same higher ups got used to it and started to perceive short, straight-to-the-point emails as "rude". So even more people are looking into using AI to generate longer emails as they don't want to seem rude. At that rate it's not long until those that rely on AI to write "better" (== longer) emails get promoted over some hard working person who simply doesn't care to write long emails.

Even though AI didn't do anything, it somehow still "trained" people to start using it and even rely on it, and it even indirectly rewards people who use it for completely meaningless and seemingly unrelated tasks. And in a few years, whatever comes out of this will be used to train the next generations of AI, creating even more of a feedback loop that people will be completely unaware of until it eventually crashes onto itself. Because that loop can't be observed in the AI itself, it has to be observed in the people.