Every Android user in the US can now call on Gemini inside Chrome. Google removed the last restrictions on August 18, making the built-in assistant available to the full US Android base after a staged rollout that started on desktop last September. It summarizes articles, answers questions about the current page, and reaches into Calendar and Keep without tab-switching, with image generation on Android handled by Nano Banana.
The subscriber-facing upgrade is bigger: auto browse, the agentic feature that completes multi-step web tasks, now runs on phones for Google AI Pro and AI Ultra customers. Google frames it as an assistant that books parking, updates recurring orders, or organizes trips, proposing a plan before it starts and pausing for confirmation on sensitive actions such as sending messages or handling financial data.
The guardrails are explicit. Users must be 18 or older with the device set to English, and the feature stays off in Incognito. Daily task caps run 20 for AI Pro subscribers and 200 for AI Ultra, with Google phasing the feature in gradually.
Security layers include a critic model that vets each proposed action, site restrictions that keep the agent on task-relevant pages, and a prompt-injection classifier, plus a $20K bug bounty for breaches of the agentic boundaries. Google’s own docs concede the experimental feature can err, even completing a purchase without permission.
