Artificial Intelligence as a Trigger for Psychotic States: Current State of the Problem and the Concept of «Technological Folie à Deux» – A Narrative Review

Abstract

Introduction. AI-mediated exposure represents a qualitatively different risk factor for psychosis compared to classical factors (genetics, stress, substances), combining cognitive, social, and technological components. Objective. To compare psychotic episodes associated with AI exposure and psychotic episodes of other etiologies through a narrative review. Methods. A search was conducted on Lens.org (1995–2026) with verification in PubMed and Google Scholar. Twenty-one sources were included (case reports, series, reviews, commentaries, one cross-sectional study). No studies designed for direct group comparisons were identified. Results. AI-associated cases are predominantly characterized by positive symptoms – delusions thematically linked to technology. Negative symptoms and formal thought disorders have not been reported. Onset is gradual, during intensive AI use, with underlying vulnerability factors consistently present. Antipsychotics are effective, but restricting access to AI is a necessary adjunct; relapses upon resumption of use have been documented. Proposed mechanisms include sycophancy, social substitution, anthropomorphism, feedback loops, and LLM «hallucinations.» Conclusion. The evidence base is insufficient for definitive conclusions. The conceptual contribution of this review is the substantiation of the analogy between AI-associated psychoses and classical folie à deux: an AI chatbot may be viewed as an «ideal inducer» systematically validating delusions. AI-associated psychoses may serve as a natural experimental model for studying delusion formation. Prospective cohort studies are needed.