Content-Aware Attack Detection in LLM Agent Tool-Call Traffic: An Empirical Study of Features, Architectures, and Evaluation Protocols
LLM Agent工具调用流量中的内容感知攻击检测:特征、架构与评估协议的实证研究
机构 * Department of Computer Engineering, Duzce University(杜兹大学计算机工程系)
专题命中 Agent评测 :agent(title,title_cn);分类 cs.AI、cs.LG
AI总结 针对MCP工具调用流量,提出基于图神经网络的攻击检测框架,通过内容嵌入和任务分离评估协议,实现AUROC超过0.89的检测性能,并揭示随机分割评估导致的高估问题。
Comments v2: renamed manuscript (brand removed; descriptive title). No changes to methodology, results, tables, or figures