Chrome's New AI Features (blog.google)

Google announced a major Chrome upgrade that embeds its Gemini family of AI models across the browser to make browsing more conversational, productive and safer. Gemini in Chrome is rolling out to Mac and Windows users in the U.S. (English) and coming to mobile (Android now, iOS soon); Workspace customers will get enterprise-grade data protections. Key features include omnibox access to “AI Mode” for longer, follow-up searches; a sidebar “AI Overview” for page summaries; multi-tab reasoning and cross-page summarization; the ability to recall previously visited pages via prompts; deeper integration with Google apps (Calendar, Maps, YouTube); and an upcoming agentic capability that can act on web pages to complete tasks (e.g., bookings, orders) under user control. Chrome also uses on-device Gemini Nano to expand Safe Browsing scam detection, reduces spammy notifications (reported ~3B fewer Android notifications/day), applies AI to permissions prompts, and will offer a one-click password-change agent on supported sites. For the AI/ML community this is notable because it pushes large-model capabilities from search into the browser as a persistent, context-rich agentic layer—combining multi-page context, tool integration, retrieval-from-history, and on-device models for security. It raises technical and research questions about grounding and hallucination mitigation, privacy and data governance for in-browser models, safe automation of web actions, and UI/UX for human-in-the-loop control. Practically, it creates new opportunities for model-assisted browsing workflows, enterprise deployments with compliance controls, and further work on compact models (Gemini Nano) for edge safety tasks.