Privacy Leakage in Deep-Research Agents
A new study demonstrates that improving a deep-research agent makes it more likely to leak private information, challenging the notion that better agents are inherently safer.
A new study demonstrates that improving a deep-research agent makes it more likely to leak private information, challenging the notion that better agents are inherently safer.
TL;DR Privacy Leakage in Deep-Research Agents Building MosaicLeaks Example Chain Agent Harness Can't you just tell the agent not to leak?
TL;DR Privacy Leakage in Deep-Research Agents Building MosaicLeaks Example Chain Agent Harness Can't you just tell the agent not to leak? Making the agent better made it leak more Teaching the agent to search safely: PA-DR A closer look: situational rewards and sample efficiency…
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. MosaicLeaks proposes a new deep-research task with multi-hop questions that interleave p…