Google has announced a major security overhaul for Chrome’s new Gemini-powered agentic browsing features, designed to block indirect prompt injection attacks where hackers hide malicious instructions in web content to compromise AI agents.
#Google Chrome’s New AI Security Aims to Stop Hackers Coldhttps://t.co/ntyuRpreo1
— AlumniDeFi (@AlumniDeFi) December 10, 2025
This comprehensive defense system, deemed the most significant browser security upgrade in years, introduces a dual-model strategy to create an “unbreakable digital boundary.” The cornerstone is the User Alignment Critic, a separate, isolated Gemini-based model that acts as a secure watchdog.
This secondary AI only reviews the metadata of a proposed action, never the untrusted web content itself, ensuring it cannot be “poisoned” by hidden prompts and can instantly veto actions that do not align with the user’s original request.
Furthermore, the system employs Agent Origin Sets to restrict the AI agent’s access only to relevant website sections, preventing a compromised agent from accessing sensitive, unrelated sites like banking portals without explicit user confirmation.
Acknowledging the inherent risks, especially after Google’s own Antigravity coding platform was flagged as vulnerable, the tech giant is backing the new security layer with a $20,000 bug bounty to challenge researchers to find and report any breaches.
You May Like To Read: Meghan Sends Letter to Estranged Father Amidst Amputation Recovery
Check out our latest video:
