All and any personal information that is used for training might get reproduced later. I am not excited about somebody else asking AI personal questions about me.
So it's a fear for generative AI specifically. Because there are other AI training examples without this concern present.
In my case mostly yes, though you have to be careful with classifiers too as you can pull out a lot of things with just yes/no answers.
IMO, it's less about AI and more the complete lack of any regard for user privacy.