AI as therapist or assistant? Opportunities and boundaries of artificial intelligence in mental health care

Purpose This study aims to examine where generative artificial intelligence (AI) may add value in mental health care, where its clinical limits remain and how human oversight should change as AI assumes greater responsibility. Design/methodology/approach This opinion piece draws on recent randomized trials, systematic reviews, safety evaluations and international guidance on conversational AI in mental health. The evidence is interpreted through a three-level framework: supportive self-management, clinician-augmented care and autonomous clinical care. Findings AI-supported interventions can improve symptoms of depression and anxiety in some settings and may widen access to low-intensity support. However, evidence remains heterogeneous and important concerns persist around suicide-risk responses, privacy, bias, over-reliance and accountability. Current evidence supports AI more strongly as an adjunct than as an autonomous therapist. Social implications Access alone should not be mistaken for inclusion. Safe mental health AI also requires attention to digital access, language, culture, affordability and routes to human care. Originality/value The paper proposes a practical clinical hierarchy in which the evidence threshold, regulatory scrutiny and human oversight increase with the clinical responsibility assumed by AI.