Paper Proposes EB-MAS to Add Auditable Belief Updates as Safety Layer for LLM Multi-Agent Systems

[arXiv] A preprint by Zhuoyun Li, Boxuan Wang, Jinwei Hu, Xiaowei Huang and Yi Dong proposes 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, according to the paper. The authors define belief as an engineering abstraction for evidence-based commitments that guide downstream action. They argue that 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. The paper also discusses group-level belief risks, including manufactured consensus, collusion and cascading misinformation, and outlines integration challenges around grounding, false positives, scalability and ethical misuse. The authors said 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. The work is a preprint and has not been peer-reviewed.