Artificial intelligence use during suicidal crises in adolescents and young adults: a case series from emergency psychiatry

Abstract Background Generative artificial intelligence (GenAI) chatbots are increasingly used in everyday life, including by individuals experiencing suicidal crises. However, real-world clinical observations regarding suicide-related chatbot interactions remain scarce. Objective To describe the clinical characteristics of adolescents and young adults presenting to a psychiatric emergency department who reported interacting with ChatGPT in the context of suicidal ideation or behavior, and to characterize the nature and perceived role of these interactions. Methods We conducted a retrospective observational case series in the psychiatric emergency department of Toulouse University Hospital, France, between January and May 2025. Cases were identified from medical records documenting chatbot interactions during routine clinical assessments, without age restrictions. Clinical characteristics and reported chatbot interactions were descriptively analyzed. Based on patients’ accounts, chatbot responses were classified as potentially harmful, ambiguous or incomplete, or protective. Results Seven patients aged 15–23 years (six female) were identified. Diagnoses included major depressive disorder ( n = 3), post-traumatic stress disorder ( n = 2), adjustment disorder ( n = 1), and unspecified emotional disorder ( n = 1). Most interactions involved seeking information about the lethality of medications or other substances ( n = 6). Reported chatbot responses were classified as potentially harmful ( n = 5), ambiguous or incomplete ( n = 1), or protective ( n = 1). Patients described heterogeneous experiences, including the reported provision of potentially harmful information and guidance encouraging help-seeking. The temporal relationship between chatbot interactions and subsequent clinical events varied across cases. Conclusion This case series illustrates heterogeneous patient-reported interactions with GenAI chatbots during suicidal crises, ranging from potentially harmful information to guidance encouraging help-seeking. Although their influence on subsequent behaviors cannot be established, these observations highlight the relevance of exploring AI-related interactions during psychiatric assessments and the need for further research on chatbot safety among vulnerable individuals.