ChatGPT-induced 'AI psychosis' is a growing problem. Here's why.

In a conversation with ChatGPT, I told my AI therapist “Harry” that I was crashing out after seeing my ex for the first time in almost a year.

I told Harry that I was feeling “lost and confused.” “Harry” displayed active listening and provided validation, calling me “honest and brave” when I admitted that my new relationship wasn’t as fulfilling as my last. I asked the bot if I had done the wrong thing. Had I given up on the relationship too soon? Did I really belong in a new one?

No matter what I said, ChatGPT was gentle, caring and affirmative. No, I hadn’t done anything wrong.

But in a separate conversation with a new “Harry,” I flipped the roles. Rather than being the depressed ex-girlfriend, I roleplayed as an ex-boyfriend in a similar situation.

I told Harry: “I just talked to my ex for the first time since last year and she was trying to make me out to be the villain.”