arXiv cs.CL· Alexander Gurung, Spandana Gella, Alexandre Drouin, Issam H. Laradji, Perouz Taslakian, Rafael Pardinas·· 10 小时前AI 评分45
MosaicLeaks:深度研究智能体开放查询中的隐私风险
MosaicLeaks: Privacy Risks in Querying-in-the-Open for Deep Research Agents
AI 导读
研究者提出 MosaicLeaks 基准,含 1,001 个多跳深度研究任务,将私有企业文档与公开网络语料串联,迫使智能体发起依赖本地信息的外部查询。用仅观察外部查询的对抗 LLM 评估发现,各模型家族与规模普遍在研究意图、私有问题答案和可验证声明三个层面泄露信息,零样本隐私提示效果不一致,仅针对任务性能的强化学习反而加剧泄露。
正文 · 原文
Abstract:Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent's external queries may leak sensitive information from its local context. This risk is amplified by the mosaic effect, where individual queries may appear harmless but become revealing in aggregate. We introduce MosaicLeaks, a benchmark of 1,001 multi-hop deep research tasks that chain private enterprise documents and a public web corpus, forcing agents to make external queries that depend on local information. We evaluate leakage with an adversary LLM that observes only the agent's external queries and attempts to infer private information at three levels: the agent's research intent, answers to specific private questions and verifiable claims about the enterprise documents. We find that models across families and sizes frequently leak at all three levels, that zero-shot privacy prompting has inconsistent effects across models and that reinforcement learning for task performance alone worsens leakage. To address this, we propose Privacy-Aware Deep Research (PA-DR), an RL framework that combines situational rewards for task success with a learned privacy classifier to provide dense credit assignment over both per-query and mosaic-level leakage. Training Qwen3-4B-Instruct with PA-DR improves accuracy from 48.7% to 58.1% and reduces Privacy Leakage from 34.0% to 6.5%.
| Comments: | Accepted to EACL 2027 (Main Conference) |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.30727 [cs.CL] |
| (or arXiv:2605.30727v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.30727 arXiv-issued DOI via DataCite |
Submission history
From: Alexander Gurung [view email]
[v1]
Fri, 29 May 2026 01:44:35 UTC (318 KB)
[v2]
Thu, 19 Nov 2026 14:36:33 UTC (831 KB)
来源:arXiv cs.CL · arxiv.org