Delusions and Harms Associated with AI Chatbot Use: Early Evidence from 185 Real-World Reports

Abstract Importance: Reports have raised concerns that generative artificial intelligence (AI) chatbots may validate or elaborate delusional beliefs, respond inappropriately to suicidal ideation, and contribute to mental health harms, but real-world data on reported harms remain limited. Objective: To characterize psychopathological features, chatbot behaviors, timing, and outcomes in first- and second-hand accounts of mental health harm linked with AI chatbot use. Design: Cross-sectional secondary analysis of deidentified online survey responses gathered between August 7, 2025, and February 2, 2026. Setting: Data were collected through a website form hosted by lived experience support group The Human Line Project, which removed direct identifiers before transferring the reports to the research team. Participants: Individuals reporting harm to their mental health linked with AI chatbot use, and family members, friends, or partners of affected individuals. Exposure: Reported use of an AI chatbot. Main Outcomes and Measures: The primary quantitative outcome was the presence of delusional beliefs, coded by paired raters with relevant clinical experience.

Introduction. Large language model (LLM) artificial intelligence (AI) chatbots are increasingly ubiquitous, with 900 million individuals per week using ChatGPT.1 However, LLMs can be sycophantic2, which can reduce prosocial behavior and promote dependence.3 In some cases, chatbots validate or elaborate delusions,4 and respond inappropriately to suicidal ideation.5 Studies have evaluated model responses to prompts simulating psychiatric deterioration and risk6–10, but most approaches to evaluating chatbot responses in mental health contexts are not informed by real-world harm data.11 In this study we aimed to explore psychopathology, chatbot behaviors, and outcomes in a pre-collected dataset of self-selected accounts of mental health harms believed to be linked to AI chatbot use.

Discussion / Conclusion. In this article we report preliminary findings from an early dataset of AI-associated mental health harms. Most cases involved ChatGPT models, potentially reflecting market share. Companionship was the most common reason for use, which is in line with the finding that isolation was common. Grandiose delusions were recorded 1.7 times as frequently as paranoid or persecutory delusions, which might reflect chatbot sycophancy, and delusions were recorded as validated by the chatbot in almost half of cases. Inter-rater reliability for presence of delusions was only moderate (Cohen’s kappa = 0.52), consistent with previous research.13 This may reflect the difficulty of determining whether delusions were present from brief, retrospective accounts that did not always describe belief conviction or reality testing. Reported loss of insight and most belief themes were more common in second-hand reports. This may partly reflect differential ascertainment: reduced insight - common during acute delusional episodes - may limit affected individuals’ recognition or reporting of delusional experiences. However, second-hand reports may also represent more severe or conspicuous cases. Psychological harm onset was concentrated in quarters 2 and 3 of 2025.