Architecting Security for Agentic Capabilities in Chrome

Google’s Chrome security team outlines a layered defense architecture to secure agentic browsing, centered on mitigating indirect prompt injection through a trusted User Alignment Critic, origin set gating, user confirmations for sensitive actions, and real-time threat detection. The post describes continuous red-teaming, automated testing, and rapid update cycles, alongside community collaboration via updated VRP guidelines. Together, these measures extend Chrome’s origin-isolation principles and introduce trusted-model patterns to reduce data exfiltration risks and prevent misaligned agent actions.