The AI Podcast Chrome Receives New AI Tools to Spot Fake Download Prompts
Dec 9, 2025
Google's new AI technology can identify and block deceptive download prompts in Chrome. This innovation tackles a significant online threat. Featuring unique agentic capabilities, the browser will roll out strategies in the coming months. The discussion touches on factors like user alignment protecting actions and models that prevent malicious URLs. Highlighting issues with user permissions, the hosts argue for a balance between safety and seamless automation.
AI Snips
Chapters
Transcript
Episode notes
Meta Critic Checks Agent Actions
- Google uses a separate Gemini-based critique model to check each agent action against the user's original goal.
- The critic sees only metadata about planned actions so it cannot be tricked by on-screen prompt injections.
Origin Sets Limit Data Exposure
- Chrome restricts agents to a limited set of readable and writable origins to reduce cross-origin data leaks.
- The agent will ignore irrelevant elements like banner ads and only interact with whitelisted iframes and page elements.
Host Moved To Brave For Ad Blocking
- Jaeden switched from Chrome to Brave because Brave blocks ads by default and avoids ad-related annoyances.
- He highlights the irony that Google's agent design explicitly ignores ads despite ads being Google's primary revenue source.