Ethical evaluation of AI-supported mental health applications

Psychiatric ethics is a uniquely significant domain within the broader field of medical ethics, addressing complex issues such as confidentiality, privacy, informed consent, involuntary admission and treatment, and the intricacies of the physician-patient relationship. As artificial intelligence (AI) becomes increasingly integrated into psychiatric practice, including electronic health records, brain imaging systems, social media platforms, mobile applications, telepsychiatry, the Internet of Things, and therapeutic chatbots, it improves efficiency by aiding in the detection of prodromal phases of illnesses, facilitating patient follow-up, and fostering personalized treatment strategies; however, it also risks deepening existing ethical dilemmas and introducing new ones. It is vital that the algorithms and AI systems underpinning these applications are designed with ethical principles at their core, prioritizing patient benefit and avoiding harm.

A three-step approach was applied: (1) a narrative/conceptual literature review across PubMed/MEDLINE, Scopus, and Web of Science; (2) ethical and clinical analysis of four widely used AI-supported mental health applications (Wysa, Youper, Amaha, Yuna) across 14 categories; and (3) semi-structured interviews with five practicing psychiatrists, analyzed using grounded theory. Findings were interpreted through a two-part ethical framework: one layer grounded in psychiatric ethics norms, the other in Beauchamp and Childress's four biomedical principles.

Application analysis showed gaps in transparency, accountability, crisis detection, and clinical accuracy. Analysis of the interviews revealed four main themes: Ontological mismatch with psychiatry, clinical role displacement, digital empathy limitations, and emotional responses to AI. These findings were structured into a conceptual framework at the normative, practical, and perceptual levels.

Despite the growing presence of AI in psychiatric care, there is a significant lack of developed ethical frameworks specifically tailored to AI-supported mental health applications, highlighting the urgent need for a standardized ethical evaluation framework that prioritizes patient well-being and autonomy. Future research should expand clinician samples, include patient and developer perspectives, and undertake longitudinal studies to examine the long-term safety of AI systems in psychiatry.