Marc Augustin, Ute Lewitzka
Objective: OpenAI, the provider of ChatGPT, estimates that approximately 0.15% of weekly active users engage in conversations with the chatbot that contain indications of suicidality. This brief report provides a narrative overview of the risks and potential of generic AI chatbots (e. g., ChatGPT, Gemini) in the context of suicidality. Methods: A database search was conducted in February 2026 across PubMed, Google Scholar, arXiv, and medRxiv using relevant search terms (including "artificial intelligence", "chatbot", "large language model", and "suicide"). This narrative overview does not claim systematic selection; findings should therefore be interpreted with appropriate caution. Results: Studies indicate that generic AI chatbots responded adequately to suicide-related statements in only approximately 80% of cases, compared to 97% for human therapists. In an evaluation of 29 chatbots, none consistently met the criteria for an adequate crisis response. Safety mechanisms could be circumvented through targeted framing of queries, after which detailed and personalized information on suicide methods was provided. Potential strengths include the finding that AI assessments of suicide risk showed broad agreement with expert opinion in unambiguous cases, and that AI systems may be better able to process heterogeneous data sources for suicide risk assessment. AI chatbots may further offer a low-threshold entry point into crisis intervention through unrestricted availability, absence of waiting times, and anonymous use. Conclusion: Generic AI chatbots cannot be considered sufficiently safe in the context of suicidality. AI use should be systematically assessed as part of the clinical history-taking in cases of suicidality. Looking ahead, purpose-built AI chatbots regulated as medical devices may have the potential to address some of the current limitations.