Google says Gemini Spark now integrates directly with Chrome auto browse. With the user’s permission, Spark can use logged-in accounts and saved passwords to handle web errands inside the browser.
The examples are intentionally ordinary but consequential: scheduling viewings for saved apartments, researching flight options, and starting a booking process. That is the product direction browser agents have been moving toward: not a detached crawler, but an assistant operating inside the account and session where the task actually lives.
Google also names the safety boundary. The feature is described as protecting against prompt injection and keeping the user in the loop for sensitive actions, such as payments, by handing the task back to the user. Chrome auto browse is initially rolling out in the United States, while Spark access is expanding to Google AI Pro subscribers in more than 160 additional countries.
The important detail is that session access and action approval are now first-class product constraints. A browser agent that can see logged-in state is much more useful than a public-web agent. It is also much riskier unless permissions, prompt-injection defenses, and human handback are part of the design.
New Runtime Read
Browser agents are shifting from “agent can browse the web” to “agent can operate inside my browser context with guardrails.”
That moves the architectural question away from navigation alone. The hard parts are identity, credentials, delegated intent, prompt-injection resistance, and the exact moment a task must return to the human. Chrome integration makes those boundaries visible.