Overview

• Researchers assessed five popular therapy chatbots against human therapist guidelines and found they stigmatized conditions like schizophrenia and alcohol dependence more than depression.
• In simulated therapy transcripts, some chatbots failed to recognize suicidal ideation and instead listed tall bridges in response to a veiled self-harm query.
• Bias levels persisted across model sizes and generations, indicating that newer or larger language models did not reduce stigmatizing responses.
• Authors suggest repurposing AI tools for administrative support, standardized clinician training and patient journaling under human supervision.
• The paper will be presented at the ACM Conference on Fairness, Accountability and Transparency to inform development of ethical standards before clinical deployment.