Trendora

Privacy-Aware Deep Research (PA-DR)

Hold

Techniques

A reinforcement-learning training method designed to improve research performance while reducing privacy leakage.

Why it's here

Placed in Hold: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 0 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.

Evidence (1)

  • 7Hugging Face Blog·6/18/2026research
    MosaicLeaks exposes privacy risks in deep research agents

    Hugging Face describes MosaicLeaks, a new benchmark for evaluating whether deep research agents leak private information when combining local documents with external web search. The paper finds that many agents expose sensitive details through query logs, and that optimizing only for task success can increase leakage; a privacy-aware RL method improves success while reducing leakage.