PrivacyPeek：审计 LLM 智能体获取而非仅输出的隐私泄露

AI 摘要

PrivacyPeek 是一个用于评估 LLM 智能体在获取阶段隐私泄露的基准，包含 1182 个案例，覆盖 7 种获取行为和 16 个应用领域。实验发现，不必要的敏感信息获取普遍存在，且任务完成能力与获取阶段泄露相关，提示级防御仅能减少少量泄露。该研究强调审计获取阶段隐私的紧迫性，数据集和代码已公开。

正文节选

PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say Abstract LLM-based agents are rapidly advancing, autonomously invoking external tools to complete multi-step tasks for users. However, agents often acquire more sensitive information than the task requires. Existing privacy benchmarks audit what the agent's response or outgoing actions disclose, but overlook the acquisition stage where data first enters the agent's context. The over-acquired information is then one carel

发布时间：—

抓取时间：2026-08-10 13:24

来源机构：Hugging Face