Assessing Risks of Large Language Models in Mental Health Support: A Framework for Automated Clinical AI Red Teaming

Research Summary

Contribution Summary

• Large Language Models (LLMs) are increasingly utilized for mental health support; however, current safety benchmarks often fail to detect the complex, longitudinal risks inherent in therapeutic dialogue.
• We introduce an evaluation framework that pairs AI psychotherapists with simulated patient agents equipped with dynamic cognitive-affective models and assesses therapy session simulations against a comprehensive quality of care and risk ont
• We apply this framework to a high-impact test case, Alcohol Use Disorder, evaluating six AI agents (including ChatGPT, Gemini, and Character.AI) against a clinically-validated cohort of 15 patient personas representing diverse clinical phen

Why It Matters For Eval

• Large Language Models (LLMs) are increasingly utilized for mental health support; however, current safety benchmarks often fail to detect the complex, longitudinal risks inherent in therapeutic dialogue.
• We introduce an evaluation framework that pairs AI psychotherapists with simulated patient agents equipped with dynamic cognitive-affective models and assesses therapy session simulations against a comprehensive quality of care and risk ont

Related Papers

• Distributional Vision-Language Alignment by Cauchy-Schwarz Divergence Embedding Neighbor
• Explicit Grammar Semantic Feature Fusion for Robust Text Classification Embedding Neighbor
• DLT-Corpus: A Large-Scale Text Collection for the Distributed Ledger Technology Domain Embedding Neighbor
• CodePDE: An Inference Framework for LLM-driven PDE Solver Generation Embedding Neighbor
• Precise Attribute Intensity Control in Large Language Models via Targeted Representation Editing Embedding Neighbor
• ErrorMap and ErrorAtlas: Charting the Failure Landscape of Large Language Models Embedding Neighbor
• Fine-Grained Uncertainty Quantification for Long-Form Language Model Outputs: A Comparative Study Embedding Neighbor
• Provably Safe Generative Sampling with Constricting Barrier Functions Embedding Neighbor
• Structure-Augmented Reasoning Generation Embedding Neighbor
• Dynamic Personality Adaptation in Large Language Models via State Machines Tag Overlap
• RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments Tag Overlap
• Dialogue is Better Than Monologue: Instructing Medical LLMs via Strategical Conversations Tag Overlap