Google Readies Gemini Spark Beta for May 19 I/O, Warning It May Make Purchases

9 articles · Updated · TestingCatalog · May 14

A Gemini web pop-up shows Google preparing “Gemini Spark BETA” for a likely public debut around the May 19 I/O keynote, recasting its agentic upgrade as an always-on assistant for inbox triage and multi-step online tasks.
Hidden onboarding text says Spark will pull from Connected Apps, chats, tasks, signed-in websites, Personal Intelligence signals and location, and may pass necessary data to third parties while acting on a user’s behalf.
Google labels the system experimental and warns it could share sensitive information or even complete purchases without asking, though it is designed to request permission before sensitive actions.
Settings controls let users clear saved remote browser data—including login details and remote code execution state—and disable Connected Apps or delete Gemini Apps Activity.
The launch would push Gemini beyond reactive chat toward a background operator across Google services, putting it more directly against emerging agent platforms from OpenAI and Anthropic.

As Google's AI gains direct control of web browsers, what new capabilities and security risks will emerge for our daily online activity?

When AI agents can make purchases without asking, how can users truly maintain control over their financial and digital lives?