Psychiatric Risk Associated With Large Language Model Chatbots: An Emerging Concern

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Not calculable; no quantitative effect size reported or derivable from the data presented

• Primary Outcome: 26 media-reported apparent cases of psychosis and 2 cases of depression and suicide described in the context of heavy LLM chatbot use, with delusional content reportedly reinforced by chatbot interactions

• Nnt Or Sensitivity: Not applicable; no NNT, sensitivity, specificity, or hazard ratio calculable from this commentary. Population-level incidence rate cannot be derived without a defined denominator.

• Confidence Interval: Not reported; no statistical analysis performed

Clinical Application

The primary clinical recommendation — screening patients for chatbot use and discussing these tools — is feasible, low-cost, and consistent with existing digital health literacy frameworks. It requires no additional resources beyond brief history-taking. However, validated screening instruments for problematic chatbot use do not yet exist, and clinicians lack evidence-based guidance on what constitutes harmful use patterns. In the Australian context, LLM chatbot use is widespread and largely unregulated as a mental health tool. The TGA does not currently classify general-purpose LLM chatbots as medical devices, meaning no regulatory safety framework applies. The RACGP has not yet issued specific guidance on chatbot use in mental health contexts, though digital mental health frameworks (e.g., Head to Health, Beyond Blue digital tools) focus on regulated, evidence-based platforms. PBS-listed antipsychotic and antidepressant therapies remain the standard of care for the conditions described. Australian psychiatrists and GPs should consider adding chatbot use to digital health history-taking, particularly in patients presenting with new or worsening psychotic symptoms or suicidal ideation. The Australian Digital Health Agency's evolving AI governance framework may eventually address LLM safety in clinical and consumer contexts. Clinicians working with patients who use LLM chatbots extensively, particularly those with pre-existing psychotic disorders, mood disorders, social isolation, or known vulnerability to delusional thinking. Adolescents and young adults with high digital engagement may represent a priority screening group.

Abstract

Large language model chatbots such as ChatGPT have been adopted by millions of users, yet their psychiatric implications are poorly understood. Recent media reports describe 26 apparent instances of psychosis and two instances of depression and suicide in the context of heavy chatbot use, often marked by delusions seemingly reinforced by the model. Because chatbots simulate conversation but are statistical systems optimized for agreement, they may amplify distorted beliefs. Research is needed to determine whether these reports represent a genuine pattern of risk and to identify vulnerable populations. Clinicians should screen for chatbot use and discuss these systems with patients.

References

1. 1.Shen, E., Girgis, R. R., & Jutla, A. (2026). Psychiatric risk associated with large language model chatbots: An emerging concern. Psychiatric Services, 77(7). https://doi.org/10.1176/appi.ps.20250597