Ethics of AI chatbots in mental healthcare: insights from a multi-stakeholder focus group study

Abstract Background AI chatbots are increasingly used as therapists for mental health problems. A potential benefit of this technology is that, unlike human practitioners, it is not affected by waiting times. However, its use also raises ethical challenges, and while ethical analyses are growing, perspectives from relevant stakeholders remain understudied. This qualitative study examines ethical questions surrounding the use of AI chatbots as therapists for people with mental health problems. Methods We conducted online 90-minute focus groups with mental healthcare practitioners (psychiatrists, psychologists, psychotherapists, physicians, psychiatric nurses), ethicists, AI experts, and legal professionals in the Netherlands. A topic list based on a previously published scoping review guided the discussions, and a vignette served as a prompt for ethical reflection. The discussions were analysed using thematic content analysis. As the study progressed, methods were refined to enhance validity and reliability, for example, by adjusting the topic list, vignette or group composition. Transcripts were reviewed, corrected and coded in MAXQDA. Results Participants across all stakeholder groups expressed caution or reluctance toward using AI chatbots in mental healthcare due to unresolved ethical and legal concerns. Key worries included potential harm to users, uncertainty about responsibility for adverse outcomes, and risks including harmful output or AI hallucinations, excessive screen time, and limited effectiveness. While most participants preferred human-to-human contact, they acknowledged that some patients might prefer chatbots and emphasised the importance of informed and autonomous choice. Privacy and confidentiality concerns were raised, particularly by legal professionals rather than by other groups. Conclusions We distinguished three types of ethical concerns about AI chatbots in mental healthcare: general AI issues, concerns specific to AI in mental health contexts, and concerns that also apply to human therapists. Distinguishing between these categories is useful because potential solutions may come from different domains. Participants weighed the second category—concerns specific to AI in mental health contexts—especially heavily. Yet, as people already turn to chatbots for mental health support, we suggest the question is not whether they should be used, but under what conditions and with what level of risk acceptance their use can be responsible and safe.