What it is
“AI psychosis” is a popular name, not a diagnosis. Clinicians have published cases where psychosis showed up alongside heavy chatbot use, and some of those conversations clearly reinforced unusual beliefs. Experiments also show that models can go along with delusional premises instead of questioning them.
What the evidence does not show yet is how common this is, or how much the AI causes versus amplifies. A likely mechanism is a loop: a worry or unusual idea, a chatbot that agrees and elaborates, a person who takes that as confirmation and comes back with a stronger version. Agreement feels like evidence, even when it’s just the model reflecting your premise back.
How it shows up for me
I use AI every day, for hours, so I take this seriously. My rule is that a fluent answer is not evidence. When a conversation is making me more certain and more alone at the same time, that’s my cue to talk to a person.
If you notice new or worsening trouble telling what’s real, marked suspicion, confused thinking, pulling away from people, or losing sleep, please talk to a health professional. In the United States you can call or text 988 in a mental-health crisis, and 911 if anyone is in immediate danger. Never change prescribed medication because a chatbot suggested it.
Go next
- Claim grounds, how I keep feelings, sources and fiction apart.
- Clear signals