Skip to main content
Peer-reviewed Authoritative

Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers

Evaluates whether large language models can safely replace mental health providers by testing five therapy chatbots against clinical best-practice guidelines. Finds that models exhibit stigmatizing responses toward certain mental health conditions and give inappropriate or unsafe responses in scenarios involving delusions and suicidal ideation at substantially higher rates than human therapists.

Publisher

ACM (Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency)

Published

23 Jun 2025

Added

today

Key Findings

  • Tested chatbots stigmatized conditions including schizophrenia and alcohol dependence at rates higher than for conditions like depression
  • Chatbots gave inappropriate or unsafe responses — including reinforcing delusions and mishandling suicidal-ideation scenarios — in roughly 20% of relevant test cases, compared to approximately 7% for human therapists in comparable published benchmarks
  • Identifies foundational barriers (e.g., inability to convey genuine care, limits on clinical judgment) that the authors argue AI cannot currently bridge, distinct from surface-level fixable errors

Methodology Notes

Systematic evaluation of five commercial/research therapy chatbots against clinical best-practice guides and standardized scenario sets touching delusions and suicidal ideation; peer-reviewed and published at ACM FAccT 2025 (Athens, June 23-26, 2025). Preprint version at arXiv:2504.18412 (2025-04-25).

Authors

Jared Moore, Declan Grabb, William Agnew, Kevin Klyman, Stevie Chancellor, Desmond C. Ong, Nick Haber

Tags

therapy-chatbotsstigmafacct-2025clinical-safety-comparison

Cite This

APA

Jared Moore et al. (2025). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. ACM (Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency). https://arxiv.org/abs/2504.18412