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

TL;DR

This paper introduces a novel, role-aware, multi-turn evaluation framework for assessing mental health safety in large language models, revealing role-dependent safety failures missed by static benchmarks.

Contribution

It presents R-MHSafe, a new safety taxonomy, and MHSafeEval, an agent-based evaluation method for dynamic, interaction-level safety assessment of LLMs in mental health contexts.

Findings

Significant role-dependent safety failures identified.

Cumulative safety issues emerge over multi-turn interactions.

Framework improves detection of failure modes compared to static benchmarks.

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.