MIRROR：面向LLM多智能体通信的多路径仲裁完整性
原标题：MIRROR: Multipath Quorum Integrity for LLM Multi-Agent Communication
AI 摘要
论文提出 MIRROR，一种面向 LLM 多智能体系统（LLM-MAS）通信层的完整性原语，用于防御 Agent-in-the-Middle（AiTM）攻击。MIRROR 将同一规范化载荷复制到多条逻辑路由，仅当严格多数路由报告相同摘要时才接受消息，其安全性完全依赖诚实路由占多数的假设，而非无密钥哈希本身。在 MMLU、HumanEval、MBPP 及 MetaGPT 生产 API 部署中，MIRROR 在阈值以下将攻击成功率降至 0%，且 LLM token 成本低于 LLM-as-a-Judge，后者在拓扑扫描中最多阻断 44.2% 的良性输出。
正文节选
Foundations of Language Model Security: Theory, Practice and Fundamental Limits MIRROR: Multipath Quorum Integrity for LLM Multi-Agent Communication Abstract Inter-agent communication is central to Large Language Model Multi-Agent Systems (LLM-MAS), but it introduces an underexplored vulnerability: Agent-in-the-Middle (AiTM) attacks that manipulate messages in transit without compromising the agents themselves. Prior work reports Attack Success Rates (ASR) approaching 100% on structured tasks. E