MHSafeEval: Role-Aware Interaction-Level Evaluation of Mental Health Safety in Large Language Models

2026-04-19

AI Summary

Current evaluation methods struggle to capture the clinical harms arising from the interactive and context-dependent nature of large language models in multi-turn psychological counseling. This work proposes R-MHSafe, a role-aware safety taxonomy, and MHSafeEval, a closed-loop agent-based evaluation framework. It introduces role perspectives—such as perpetrator or instigator—to characterize harm types, integrating clinical psychology knowledge to construct a dual-dimensional role–harm schema. By simulating adversarial multi-turn agent interactions, the approach enables trajectory-level safety analysis. Experimental results demonstrate that this method substantially expands coverage of failure modes and enhances diagnostic granularity, uncovering role-dependent and cumulative risks that existing benchmarks commonly overlook.

Technology Category

• Natural Language Processing: Safety and Robustness
• Multiagent Systems: Adversarial Agents
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Security and Privacy: Large-scale security measurements

Abstract

Large language models (LLMs) are increasingly explored as scalable tools for mental health counseling, yet evaluating their safety remains challenging due to the interactional and context-dependent nature of clinical harm. Existing evaluation frameworks predominantly assess isolated responses using coarse-grained taxonomies or static datasets, limiting their ability to diagnose how harms emerge and accumulate over multi-turn counseling interactions. In this work, we introduce R-MHSafe, a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of the interactional roles an AI counselor adopts, including perpetrator, instigator, facilitator, or enabler, combined with clinically grounded harm categories. Then, we propose MHSafeEval, a closed-loop, agent-based evaluation framework that formulates safety assessment as trajectory-level discovery of harm through adversarial multi-turn interactions, guided by role-aware modeling. Using R-MHSafe and MHSafeEval, we conduct a large-scale evaluation across state-of-the-art LLMs. Our results reveal substantial role-dependent and cumulative safety failures that are systematically missed by existing static benchmarks, and show that our framework significantly improves failure-mode coverage and diagnostic granularity.

Problem

• mental health safety
• large language models
• interaction-level evaluation
• clinical harm
• multi-turn counseling

Innovation

• role-aware evaluation
• mental health safety
• multi-turn interaction
• adversarial agent-based testing
• harm taxonomy