GUARDIAN: Safeguarding LLM Multi-Agent Collaborations with Temporal Graph Modeling

TL;DR

GUARDIAN is a novel framework that models multi-agent LLM interactions as temporal graphs to detect and mitigate safety issues like hallucinations and errors with high precision.

Contribution

It introduces a graph-based approach with an unsupervised encoder-decoder and information bottleneck to effectively identify safety concerns in multi-agent LLM collaborations.

Findings

Achieves state-of-the-art accuracy in safety detection.

Effectively models propagation of hallucinations and errors.

Demonstrates efficient resource utilization.

Abstract

The emergence of large language models (LLMs) enables the development of intelligent agents capable of engaging in complex and multi-turn dialogues. However, multi-agent collaboration faces critical safety challenges, such as hallucination amplification and error injection and propagation. This paper presents GUARDIAN, a unified method for detecting and mitigating multiple safety concerns in GUARDing Intelligent Agent collaboratioNs. By modeling the multi-agent collaboration process as a discrete-time temporal attributed graph, GUARDIAN explicitly captures the propagation dynamics of hallucinations and errors. The unsupervised encoder-decoder architecture incorporating an incremental training paradigm learns to reconstruct node attributes and graph structures from latent embeddings, enabling the identification of anomalous nodes and edges with unparalleled precision. Moreover, we introduce a graph abstraction mechanism based on the Information Bottleneck Theory, which compresses temporal interaction graphs while preserving essential patterns. Extensive experiments demonstrate GUARDIAN's effectiveness in safeguarding LLM multi-agent collaborations against diverse safety vulnerabilities, achieving state-of-the-art accuracy with efficient resource utilization. The code is available at https://github.com/JialongZhou666/GUARDIAN