arXiv cs.AI· Shaswata Mitra, Raj Patel, Subash Neupane, Sudip Mittal, Md Rayhanur Rahman, Shahram Rahimi·· 10 天前AI 评分31
规则止于何处，裁判始于何处：测量多智能体系统安全中的判断边界
Where Rules End and Judges Begin: Measuring the Judgment Boundary in Multi-Agent Systems Security
AI 导读
研究将 LLM 多智能体系统（MAS）的防御组织为五项原则并实现为 DEFER1，包含 28 项检查级联，能拦截的拦截、其余交给四人裁判小组。在四个领域的独立测试中，攻击成功率从约 30.0% 降至约 3.0%，78% 被拦截的攻击由确定性检查处理；在安全运营领域仅四分之一提案到达裁判。风险评分审批门会错误批准多数攻击提案、却只批准少数合法提案。
正文 · 原文
Abstract:LLM-based multi-agent systems (MAS) engage tools, share memory, and delegate tasks, often encountering adversarial content. Current defenses for MAS are typically evaluated in isolation, focusing on one attack type at a time, which can lead to costly and hard-to-audit outcomes. This study organizes defenses into five principles, implementing them as DEFER1 (DEterministic-First Enforcement with Residual judgment), which includes a cascade of 28 checks that blocks what it can and refers the rest to a panel of four judges. In independent testing across four domains, attack success rates drop from about 30.0% to approximately 3.0%, with 78% of blocked attacks handled by deterministic checks. Only a quarter of proposals reach the judges in the security-operations domain, illustrating that the rules provide security for attacks violating clear policies, while judges manage those that only misrepresent intent. Both systems have weaknesses, such as a risk-score approval gate that inaccurately approves most attack proposals but few legitimate ones, highlighting the challenges in assessing threats accurately.
| Comments: | 26 pages, 20 figures, 24 tables |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2610.07657 [cs.AI] |
| (or arXiv:2610.07657v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.07657 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shaswata Mitra [view email]
[v1]
Tue, 6 Oct 2026 02:52:51 UTC (255 KB)
来源：arXiv cs.AI · arxiv.org