ChatGPT: OpenAI, Sam Altman, AI, Joe Rogan, Artificial Intelligence, Practical AI Google Equips Chrome With Continuous AI Defense Scanning
Dec 9, 2025
Google Chrome introduces always-on AI scanning to tackle evolving cyberthreats, improving anomaly detection. The discussion highlights Chrome's agentic features and the importance of browsers as key agents. There's an interesting breakdown of Google’s Gemini critiquing model, which ensures alignment with user goals. The podcast dives into iframe risks and how these AI agents can effectively mitigate phishing threats. User permissions for sensitive actions are also examined, weighing usability against security.
AI Snips
Chapters
Transcript
Episode notes
Agent Critic Enforces Original Goals
- Google uses a separate
Critic Model Sees Goals, Not Screen Content
- Google runs a 'user alignment critique' that only sees the user's goal and planned actions, not screen content.
- If an action diverges the critic asks the planner to rethink, limiting prompt-injection attacks.
Origin Restrictions Reduce Data Leaks
- Google restricts which page origins agents can read and write to reduce cross-origin leaks.
- The browser enforces readable and writable origin sets so models only get relevant data.