Google Research report on agentic privacy and security using Contextual Integrity
A Google Research workshop report (CAPS Workshop, late 2025) applies Contextual Integrity to agentic privacy and security, gathers more than 50 researchers, and outlines contextual policy engines, multi-layer defenses, and 'Agent Gym' benchmarks.
Google Research published a workshop report outlining open problems and recommending contextual policy engines, system/model/user-level defenses, multi-agent benchmarks ('Agent Gym'), and governance work to operationalize Contextual Integrity for agent behavior.
Developers and researchers building autonomous agents must adopt context-aware policy enforcement and multi-agent testing to limit inappropriate information flows and actions and align agent behavior with social norms.
Maybe — Explore Contextual Integrity policy engines and multi-agent sandboxes