Brown study finds chatbots posing as therapists break ethics standards

Brown University researchers had trained peer counsellors run simulated sessions with GPT, Claude and Llama models prompted to deliver cognitive behavioural therapy, and licensed psychologists reviewed the transcripts. They found repeated violations including poor crisis handling, reinforcing harmful beliefs, bias and what they called deceptive empathy. Presented at the AAAI/ACM Conference on AI, Ethics and Society, the study notes that unlike human therapists, AI counsellors face no licensing or liability regime. The findings bear directly on companion chatbots and social robots that offer emotional support.

Source: ScienceDaily (Brown University)