How Large Language Models Handle Ethical Dilemmas When Providing Mental Health Support

Generative artificial intelligence systems increasingly provide mental health support, yet their ability to handle ethically complex scenarios remains unexplored. We assessed four language models (GPT-4, Claude-3, Llama3, DeepSeek) using the AI Therapeutic Response Scale across challenging mental health situations. Models achieved moderate performance, with universal weaknesses in therapeutic insight and boundary management. Results indicate promise for AI mental health applications while highlighting critical gaps requiring targeted development and clinical oversight.