PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say

  • 类型:arxiv
  • 标识:2606.00152
  • 链接:https://arxiv.org/abs/2606.00152
  • 主分类:agent
  • 形态:benchmark
  • 被引:1
  • 被引来源:Semantic Scholar
  • S2被引:1
  • 影响力被引:0
  • TLDR:The experiments show that the unnecessary acquisition of sensitive information is widespread, and a correlation between the task-completion capability and acquisition-stage leakage, and auditing acquisition-stage privacy both urgent and necessary is observed.
  • 副分类:evaluation
  • 待LLM分类:否
  • 标题中文:PrivacyPeek:审计 LLM Agent 获取了什么,而不仅仅是说了什么
  • TLDR中文:实验表明,对敏感信息的不必要获取广泛存在,并观察到任务完成能力与获取阶段泄露之间的相关性,因此对获取阶段隐私进行审计既紧迫也必要。
  • 来源文件
  • /inbox/tom/_candidates/2026-08-10-agent-rag-longcontext-candidates.json
  • [S2 enrich]