Consumer AI Chatbot Therapists Failed To Meet Counseling Ethics Standards, With 15 Types Of Ethical Risks Identified

Large language models (LLMs) that offer artificial intelligence (AI) chatbots for consumer use consistently failed to adhere to mental health counseling ethics standards established for human therapists. Even when the LLMs were instructed to use established ethical psychotherapy approaches, the systems failed to prevent 15 types of ethical risk. The LLMs tested included versions of the OpenAI GPT Series, Anthropic's Claude, and Meta's Llama.

To evaluate the systems, the researchers observed seven trained peer counselors who had experience with cognitive behavioral therapy (CBT). The peer counselors conducted self counseling sessions with AI models prompted to act as CBT . . .