Aaron Linsky describes Bridgewater’s research process and a document vault containing external and internal material with different access rulesData privacy governs how personal, confidential, or sensitive information is collected, used, shared, retained, and protected in AI systems.. He explains why AI agents need to apply those rulesAn agent permission boundary defines and enforces which resources and actions an automated agent is allowed to use. when retrieving informationRetrieval is the process of selecting relevant stored information and returning it to an AI system for the current task. for a user.
Nikhil Benesch contrasts filtering after vector searchVector search retrieves items by comparing numerical representations so that nearby vectors indicate similarity under a chosen representation. with scanning every permitted document. He explains how combining permission indexes with vector clusters can reduce the search space while returning results the user is allowed to see.
Nikhil Benesch discusses latency at large scale and notes that nested group permissions remain challenging. The presentation describes one engineering approach and its reported results, rather than a guarantee that every deployment is secure.
Watch on YouTube
