#Side-Channel Attack - AI Papers Daily

#Side-Channel Attack

1 篇

Known By Their Actions: Fingerprinting LLM Browser Agents via UI Traces

基于 UI 行为痕迹的 LLM 浏览器智能体指纹识别

🧭 Agent ★ 深度解读

University of Oxford

仅通过页面动作时序,即可识别浏览器智能体的底层模型,F1 最高 96%,随机延迟防御可被自适应分类器破解

Agent | 2026-05-14 | cs.HC | score 9/10

LLM Agent Browser Fingerprinting Behavioral Trace Side-Channel Attack XGBoost