Explicit belief updates add safety layer for lLM multi-agent systems

Auditable Belief Updates as a Missing Safety Layer for LLM-Powered Multi-Agent Systems

Artificial Intelligence

Summary

Large language models are being used in systems where multiple AI agents work together, but safety issues arise not just from bad outputs but from how these agents change and share their knowledge. The authors define a way to represent beliefs explicitly so that systems can track and control how they update their knowledge based on evidence. They introduce a system that monitors these belief changes and blocks updates that seem weakly supported but triggered by vulnerabilities. They also warn about group risks like false consensus and misinformation spreading among agents. This work suggests that improving safety in multi-agent AI needs tools to trace and control belief changes, not just better filters or models.

What this means in practice

  • For multi-agent system developers: Add explicit belief tracking and update controls to multi-agent AI systems to reduce risks of false information spreading and unsafe decisions.
  • For cybersecurity teams: Incorporate belief audit mechanisms to detect and block potentially harmful belief updates triggered by attack vulnerabilities in AI-powered cooperative tools.

A position paper. It proposes an approach and reports no results.

Authors

Zhuoyun Li, Boxuan Wang, Jinwei Hu, Xiaowei Huang, Yi Dong

Abstract

Large Language Models (LLMs) are increasingly being used as cognitive interfaces that shape how users interpret evidence, communicate, and make decisions. As deployed systems become more multimodal, tool-using, and multi-agent, safety failures no longer arise only from isolated harmful outputs, but also from how agents update, store, and propagate epistemic commitments across retrieval, memory, tools, and peer communication. In this paper, we define belief as an engineering abstraction for evidence-based commitments that guide downstream action. Building on this view, we propose the Explicit Belief Multi-Agent System (EB-MAS), which externalizes belief states into a structured substrate, uses belief operators to propose candidate updates, and applies a Cognitive Firewall to bound or reject updates that are weakly evidenced but strongly driven by vulnerability cues. We also discuss the risk of group belief at the group level, including manufactured consensus, collusion, and cascading misinformation, and outline practical integration challenges around grounding, false positives, scalability, and ethical misuse. Therefore, scalable agent safety requires not only better output filtering or stronger base models, but also explicit interfaces for tracing, calibrating, and constraining belief revision at runtime.
View PDFOpen arXiv