For me at least it's not security per se. There are plenty of unsecure things. I email all kinds of private things to my therapist. For the most part, I don't think anyone cares enough to go digging through my email correspondence.

It's two things. Firstly, I don't want an actual audio recording of my session floating around out there unless it was under my control. I don't know, but that feels like too much. If the therapist doesn't want to write notes, why can't they record themselves giving a summary of the session and let AI summarize that.

Second, I don't think every word I utter is important enough to record for posterity. Maybe I said something that I regret saying. A real therapist might have the judgement to know I didn't mean something literally or I'm dissociated and something came out wrong. Maybe I'm angry and said something rude about someone. Maybe I wished harm upon someone. Does AI understand that I'm obviously not going to actually harm anyone, that I'm just voicing my anger and frustration? Any number of things that there are nuances to that I don't think an AI is going to understand or comprehend.