You can't easily apply natural selection to social topics. Also, even staying in that mindframe: Being vulnerable to AI psychosis doesn't seem to be much of a selection pressure, because people usually don't die from it, and can have children before it shows, and also with it. Non-AI psychosis also still exists after thousands of years.

Even if AI psychosis doesn’t present selection pressure (I don’t think there’s a way to know a priori), I highly doubt it presents an existential risk to the human gene pool. Do you think it does?