AI Breakdown Chrome Gets AI Tools to Fight Modern Phishing Threats
Dec 9, 2025
Discover how Google’s AI-powered tools are transforming Chrome to combat phishing threats. The new system detects suspicious activity in real-time, significantly enhancing user safety. Learn about its innovative features, like a critic model that aligns actions with users' goals and restrictions to limit data access. Jaeden emphasizes the importance of user permissions for sensitive tasks and highlights Chrome's ability to read page code, making it easier to identify malicious sites. Tune in for insights that could shape the future of online security!
AI Snips
Chapters
Transcript
Episode notes
Metadata Critic Keeps Agents Aligned
- Google uses a separate critic model to check an agent's planned actions against the user's original goal.
- The critic only sees metadata, preventing prompt-injection via visible web content and keeping actions aligned.
Origin Sets Limit Data Exposure
- Chrome restricts agent access using readable and writable origin sets to limit which page content models can consume.
- This separation bounds cross-origin data leaks by preventing irrelevant elements (like ads) from reaching the model.
Switched From Chrome To Brave For Ads
- Jaeden recounts moving from Chrome to Brave because Chrome made ad blockers harder to use.
- He appreciates Brave's default ad-blocking and says he often forgets ads even exist online.