Attention Knows Whom to Trust: Attention-based Trust Management for LLM Multi-Agent Systems

2025-06-03

AI Summary

LLM-based multi-agent systems (LLM-MAS) suffer from robustness deficiencies due to blind trust in input messages; existing works address isolated credibility threats in a fragmented manner, lacking systematic credibility modeling. Method: We propose an attention-based message credibility assessment framework: for the first time, we analyze LLM internal attention head behaviors across six orthogonal trust dimensions, revealing that specific heads intrinsically detect distinct categories of trust violations—enabling lightweight, prompt-free, verifier-free, and training-free credibility inference. We further design a dual-granularity trust management system (TMS) operating at both message- and agent-level. Contribution/Results: Extensive multi-task experiments demonstrate significant improvements in system robustness against adversarial and noisy inputs. The approach is modular, plug-and-play, and requires no architectural or training modifications to underlying LLMs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large Language Model-based Multi-Agent Systems (LLM-MAS) have demonstrated strong capabilities in solving complex tasks but remain vulnerable when agents receive unreliable messages. This vulnerability stems from a fundamental gap: LLM agents treat all incoming messages equally without evaluating their trustworthiness. While some existing studies approach the trustworthiness, they focus on a single type of harmfulness rather than analyze it in a holistic approach from multiple trustworthiness perspectives. In this work, we propose Attention Trust Score (A-Trust), a lightweight, attention-based method for evaluating message trustworthiness. Inspired by human communication literature[1], through systematically analyzing attention behaviors across six orthogonal trust dimensions, we find that certain attention heads in the LLM specialize in detecting specific types of violations. Leveraging these insights, A-Trust directly infers trustworthiness from internal attention patterns without requiring external prompts or verifiers. Building upon A-Trust, we develop a principled and efficient trust management system (TMS) for LLM-MAS, enabling both message-level and agent-level trust assessment. Experiments across diverse multi-agent settings and tasks demonstrate that applying our TMS significantly enhances robustness against malicious inputs.

Problem

• LLM agents lack trust evaluation for incoming messages
• Existing studies ignore holistic trustworthiness from multiple perspectives
• Need lightweight method to assess message trustworthiness in multi-agent systems

Innovation

• Attention Trust Score (A-Trust) evaluates message trustworthiness
• Analyzes attention behaviors across six trust dimensions
• Enables message-level and agent-level trust assessment