The paper introduces MAS-Guard, a unified framework for safeguarding single- and multi-agent AI applications against threats such as unsafe actions and information leakage.
The paper introduces a unified framework for safeguarding single- and multi-agent AI applications, which is novel compared to existing solutions that address individual threats.
Before reading this…
Applications
To understand this paper, make sure you know these concepts first:
Agentic AI extends large language models (LLMs) from generating responses to planning and executing complex workflows, substantially expanding their utility while exposing applications to unsafe actions, information leakage, and adversarial manipulation. Multi-agent systems further expand these capabilities through specialization and collaboration, but distributed authority, heterogeneous roles, and opaque or potentially compromised participants make safeguarding even more challenging. Existing defenses address important individual threats but provide fragmented coverage across local execution, system-level interactions, and enforcement mechanisms. We introduce MAS-Guard, a unified framework that closes this gap for safeguarding single- and multi-agent AI applications containing both instrumentable and opaque agents. Within instrumentable agents, the proposed local Agent Guardrail (A-Guard) layer applies task-specific experts to assess agentic events. At the system level, developer-owned Agent Charters and a Control Plane combine deterministic structural checks of routes, agent capabilities, and message formats with semantic assessments of roles and message content. These complementary layers enable enforcement through a shared set of interventions: allow, reflection, confirmation, and block. Evaluated across six benchmarks covering diverse single- and multi-agent scenarios with three agent LLMs from different providers, MAS-Guard reduces the overall attack success rate by 71.5-86.3% relative to unguarded agents, while meeting the utility reference criterion on benign tasks.
Does Multi-Agent Collaboration Pay When Agents Serve Users? Testing Agent Scaling in Dual-Control Se…
This paper tests the effectiveness of multi-agent systems (MAS) compared to sing…
Cryptographically Secure Agent-to-Agent Communication in Multi-Agent Systems
This paper presents a zero-trust architecture for securing multi-agent system co…
Collective Emergent Alignment: Measuring and Amplifying Interaction-Induced Reasoning Alignment in M…
This paper investigates whether multi-agent large language models can exhibit em…
PersonalPlan: Planning Multi-Agent Systems for Personalized Programming Learning
The paper introduces MAP-PPL, a profile-conditioned multi-agent planning dataset…
EARS: Explanatory Abstention for Reliable Sub-Agent Modeling in Large-scale Multi-Agent Systems
EARS is a framework that improves the reliability of sub-agents in centralized m…
Skill-MAS: Evolving Meta-Skill for Automatic Multi-Agent Systems
This paper proposes a novel method called Skill-MAS that decouples experience re…
Learning Coordinated Preference for Multi-Objective Multi-Agent Reinforcement Learning
This paper proposes a new method for cooperative multi-objective multi-agent rei…
Enhancing Decision-Making with Large Language Models through Multi-Agent Fictitious Play
This paper proposes a novel multi-agent system paradigm called Multi-Agent Ficti…