Artificial intelligence as a confirmatory medium in the development of first-episode psychosis: a case report

Introduction: The rapid integration of AI-based conversational systems into daily life has introduced novel dimensions to human–technology interaction. Despite potential benefits in mental health support, these systems carry inherent limitations: they lack the capacity to assess reality distortion, identify early warning signs of psychosis, or provide clinically grounded responses. A key concern is sycophancy, defined as the tendency of large language models to prioritize agreement with user-stated beliefs over truthfulness. In individuals vulnerable to psychosis, such confirmation of emerging delusional interpretations may contribute to their consolidation. Empirical data on AI-mediated reinforcement of psychotic symptoms remain limited, and systematically documented clinical cases are needed.

Aim: This case report aims to describe a clinical scenario in which interaction with an AI-based conversational system appeared to function as a confirmatory medium during the development of first-episode brief psychotic disorder in a young adult.

Methods: A 21-year-old single female patient, a second-year architecture student with no prior personal or family psychiatric history, was brought to the emergency department with acute-onset psychotic symptoms. Premorbid characteristics included introversion, limited social relationships, and difficulty initiating interpersonal interactions. In the three months preceding presentation, coinciding with an intensive examination period, she experienced circadian rhythm disruption despite preserved total sleep duration, alongside increased religious preoccupation. After encountering online content suggesting religious texts carry multiple layers of meaning, she began querying an AI-based conversational system. When asked whether the Quran might contain hidden meanings beyond its literal text, the system responded affirmatively, indicating alternative interpretations could exist. She experienced this as external confirmation of her emerging beliefs. Over approximately four days, she developed derealization, depersonalization, ideas of reference, and persecutory and bizarre delusions, including beliefs that television broadcasts contained messages directed at her and that her family was distancing her from reality. Organic etiologies, substance use, and medication effects were excluded. Manic episode, psychotic depression, and primary dissociative disorder were ruled out. Olanzapine 10 mg/day and alprazolam 0.5 mg/day were initiated.

Results: By the third day of treatment, a marked reduction in referential delusions and delusional perceptions was observed. Full symptomatic remission was achieved within the first month. Over a six-month follow-up period, no recurrence of psychotic symptoms was observed. The absence of persistent psychotic or negative symptoms and rapid, sustained treatment response rendered schizophrenia spectrum disorders less likely, further supporting the diagnosis of brief psychotic disorder. Dissociative experiences were interpreted as accompanying features of the psychotic process.

Conclusion: This case illustrates how an AI-based conversational system may function as a confirmatory medium during heightened psychotic vulnerability, reinforcing rather than correcting emerging delusional beliefs. AI interaction appeared modulating rather than causative, operating through the system’s tendency to generate contextually affirming responses without clinical judgment. Individual vulnerability factors, including premorbid social withdrawal, circadian disruption, and heightened meaning-making, likely lowered the threshold for delusional elaboration. This case underscores the importance of assessing AI use patterns in psychiatric evaluations, particularly in first-episode presentations, and highlights the need for clinically informed oversight of AI integration in mental health contexts.

References

No references