TrustMH-Bench: A Comprehensive Benchmark for Evaluating the Trustworthiness of Large Language Models in Mental Health

Researchers have developed TrustMH-Bench, a comprehensive framework to evaluate the trustworthiness of Large Language Models (LLMs) in mental health applications. Testing revealed that both general-purpose and specialized mental health LLMs, including advanced models like GPT-5.1, significantly underperform across critical trustworthiness dimensions in mental health scenarios.

• →TrustMH-Bench evaluates LLMs across eight core pillars including reliability, crisis identification, safety, fairness, privacy, robustness, anti-sycophancy, and ethics.
• →Extensive testing of six general-purpose and six specialized mental health LLMs revealed significant deficiencies across various trustworthiness dimensions.
• →Even advanced models like GPT-5.1 fail to maintain consistently high performance across all trustworthiness dimensions in mental health contexts.
• →The research highlights critical gaps in current LLM capabilities for high-stakes, safety-sensitive mental health applications.
• →The framework establishes a systematic approach to quantify trustworthiness specifically for mental health AI applications.