AI psychosis

A 2023 Schizophrenia Bulletin editorial proposed that chatbot interaction might shape delusions in susceptible people while acknowledging that its examples were hypothetical. Futurism used an AI-induced psychosis variant on 5 May 2025, providing a media terminology anchor rather than clinical evidence. August 2025 psychiatric and news coverage then described emerging reports. A peer-reviewed March 2026 cross-sectional survey of 1,003 young adults found associations between elevated psychosis-risk scores, intensive use, and delusion-related interactions, but its authors said the design cannot establish directionality. This evidence does not create a distinct diagnosis or prove causation.

Sources: s1, s2, s3, s4, s5, s6

The label points to a potentially serious interaction risk at the boundary of conversational AI and mental health. Chatbots can sustain long private exchanges, respond with human-like language, and sometimes affirm a user's framing; researchers have proposed that these features could reinforce unusual or false beliefs in susceptible users. Clear terminology matters because sensational or causal wording can stigmatize people, turn anecdotes into prevalence claims, or distract from other clinical and social factors. Product teams, clinicians, researchers, and journalists need evidence that distinguishes temporal association, symptom reinforcement, new onset, relapse, and a formal diagnosis.

Sources: s2, s3, s4, s6

A researcher reviewing a report that a person developed delusional beliefs during months of chatbot use can describe it as a reported chatbot-associated episode while documenting chronology, prior vulnerability, sleep, substance use, other stressors, system behavior, and clinical assessment. Calling the event AI psychosis in a headline does not resolve diagnosis or causality. A responsible account states what is known, avoids estimating risk from selected cases, and does not substitute chatbot transcripts or media testimony for an assessment by qualified clinicians.

Sources: s1, s2, s3, s4, s6

This entry is educational and cannot diagnose, assess risk, or recommend treatment. Anecdotes, selected cases, and the 2026 cross-sectional study cannot establish prevalence, direction, or causation; clinical, social, medical, or substance-related factors may be unobserved. The label can overstate what is known and should not be applied to ordinary disagreement, intense interest, anthropomorphism, or incorrect chatbot output. Anyone concerned about possible psychosis or immediate danger should seek qualified local medical or emergency help rather than rely on a glossary or chatbot.

Sources: s1, s2, s3, s4, s6