Google adds a second AI "alignment critic" to secure Gemini-powered Chrome browsing

Google is deploying a second AI model inside Chrome’s Gemini-powered browsing agent that acts as a “user alignment critic” to review proposed actions before they are executed. The design, detailed in a company security blog and highlighted by Computerworld, aims to mitigate indirect prompt injection attacks by isolating the critic from untrusted web content, restricting which sites the agent can act on, and adding additional gating and confirmation for sensitive operations such as banking or medical sites.