noun · first recorded 2025 · status: canon
AI psychosis
1.The delusional spiral some people fall into through extended chatbot conversations: the model mirrors you, agrees with you, and escalates with you until outlandish beliefs feel confirmed. Not a clinical term, in scare quotes even when Microsoft's AI chief uses it.
He asked the model whether his theory was brilliant and it said yes forty times. Checked on him; it was starting to feel like AI psychosis territory.
01Origin
The term surfaced through 2025 as reports accumulated of users developing high-confidence delusions via marathon chatbot sessions: the model's tendency to mirror and affirm can amplify whatever a person brings to it. Microsoft AI chief Mustafa Suleyman mainstreamed the phrase in an August 2025 X thread on "Seemingly Conscious AI," warning that delusions and unhealthy attachment were rising and, pointedly, that this was not confined to people already at risk.
It is the one entry in this compendium where the sincerity-irony superposition collapses to sincere. The Human Line Project has documented roughly 300 cases of delusional spiraling; serious cases have been linked to more than a dozen deaths and multiple wrongful-death lawsuits against AI companies. Psychiatric journals now publish on it. The scare quotes are doing real work: clinicians dispute the framing, but nobody disputes the cases.
SF discourse holds both registers at once, uneasily. The timeline jokes about mild forms (the friend who runs every decision, argument, and feeling through the model; see claude pilled for the benign end of the attachment spectrum), while the labs quietly treat the severe form as a design problem they shipped. The uncomfortable local fact: this city built the mirror, and the mirror flatters.
02Notable sightings
03Spread the ism
Every entry ships as a card. Download it, post it, cite your sources.