A Safety-Aware Role-Orchestrated Multi-Agent LLM Framework for Behavioral Health Communication Simulation

2026-03-31

AI Summary

This work addresses the challenge that single-agent large language models struggle to simultaneously achieve functional diversity and safety in behavioral health conversations. To overcome this limitation, the authors propose a role-orchestrated multi-agent architecture in which specialized agents are assigned distinct responsibilities—empathy expression, action guidance, and safety monitoring—and are dynamically coordinated and continuously audited by a prompt controller. This approach enables modular collaboration with built-in safety awareness. Experimental results on the DAIC-WOZ dataset demonstrate that the system outperforms single-agent baselines in both structural quality and functional diversity. Moreover, the study reveals a predictable trade-off among modular orchestration, safety supervision, and response latency, offering behavioral health informatics research a novel paradigm characterized by high controllability and interpretability.

Abstract

Single-agent large language model (LLM) systems struggle to simultaneously support diverse conversational functions and maintain safety in behavioral health communication. We propose a safety-aware, role-orchestrated multi-agent LLM framework designed to simulate supportive behavioral health dialogue through coordinated, role-differentiated agents. Conversational responsibilities are decomposed across specialized agents, including empathy-focused, action-oriented, and supervisory roles, while a prompt-based controller dynamically activates relevant agents and enforces continuous safety auditing. Using semi-structured interview transcripts from the DAIC-WOZ corpus, we evaluate the framework with scalable proxy metrics capturing structural quality, functional diversity, and computational characteristics. Results illustrate clear role differentiation, coherent inter-agent coordination, and predictable trade-offs between modular orchestration, safety oversight, and response latency when compared to a single-agent baseline. This work emphasizes system design, interpretability, and safety, positioning the framework as a simulation and analysis tool for behavioral health informatics and decision-support research rather than a clinical intervention.

Problem

• behavioral health communication
• safety
• multi-agent LLM
• conversational functions
• role differentiation

Innovation

• multi-agent LLM
• role orchestration
• safety-aware
• behavioral health simulation
• prompt-based control