Google's Gemini Spark beta autonomously manages your inbox, books travel, and can make purchases, with explicit warnings about sharing sensitive data.
A leaked onboarding screen from inside the Gemini web app has revealed a product called Gemini Spark, an always-on artificial intelligence agent built to handle tasks in the background without waiting for user input. The disclosure surfaces five days before Google I/O 2026 opens on May 19.
X user Fandu spotted the screen, with the find later reported by Android Authority. The beta onboarding copy describes an agent that draws on linked apps, chat logs, scheduled tasks, location data, and something Google calls Personal Intelligence to complete multi-step workflows across both Google services and third-party platforms.
The pitch is direct: Spark handles your inbox, finishes online forms, and books travel without a single issued command.
What the agent can do on its own
The onboarding text is unusually candid about autonomous behavior. According to the leaked screen, Spark can share sensitive information with outside parties and complete purchases without pausing for confirmation in certain situations. It also retains remote browser session data, including login credentials. Disclosing these risks in the very first screen a user sees suggests Google's legal and trust teams have been thinking carefully about liability before any public launch.
Mashable's preview of the event puts a major Gemini model update at the center of the keynote. Spark would be a parallel track: not a new model, but a behavioral layer that transforms an existing model into something that acts on your behalf around the clock.
Google has prototyped this territory before. An internal project called Remy explored similar agent ideas, and existing Gemini Agent features remain locked behind the paid AI Ultra tier. Spark appears aimed at a wider consumer audience, which would mark a meaningful shift in how Google deploys autonomous artificial intelligence at scale.
The competitive backdrop
OpenAI closed a $122 billion funding round on May 13, placing its post-money valuation at $852 billion. The company reports $2 billion in monthly revenue and expects to reach one billion weekly active users faster than any platform before it. That capital base sustains aggressive product development; OpenAI's own agent work, including task automation inside ChatGPT, has been expanding since late 2025.
CNBC reported in March that OpenAI plans to nearly double its headcount to 8,000 by year-end, with most hires going into product, engineering, research, and sales. The internal alert Sam Altman reportedly issued in December 2025, triggered specifically by Google's Gemini 3, shows how closely each company tracks the other's moves.
Anthropic is moving on a different vector. Bloomberg Law reported this week that the company unveiled 12 Claude plugins covering specific legal practice areas, including corporate, regulatory, and employment law. Rather than compete on broad consumer reach, Anthropic is positioning Claude as infrastructure for high-stakes professional work, a strategy that contrasts sharply with the sweeping agentic play Gemini Spark represents.
What the timing tells you
Consumer-facing AI agents have been promised repeatedly over the past two years, and most have arrived with friction: incomplete integrations, ambiguous privacy terms, and reliability problems that erode trust faster than any onboarding screen can build it. Google surfacing Spark's risks in the first screen a user encounters is either a sign of regulatory caution or a deliberate effort to manage expectations, and is likely both.
Leaks before a developer conference sharpen anticipation and shape press coverage before the keynote begins. Whether Spark ships as a finished product on May 19 or lands as a limited extended beta remains unknown; Google has not confirmed the product exists.
The harder question is how users will calibrate trust when an agent can quietly spend money or share account data on their behalf. Convenience sells. Liability is harder to put in an onboarding screen.
Frequently asked questions
What is Gemini Spark?
A beta AI agent from Google that operates continuously in the background, completing tasks like booking flights, managing email, and running multi-step workflows without requiring manual prompts from the user.
How is Spark different from existing Gemini features?
Current Gemini Agent tools are restricted to paid AI Ultra subscribers. Spark appears to target a broader consumer audience and is designed to act autonomously rather than respond to individual queries one at a time.
Can Gemini Spark make purchases without asking?
According to the leaked onboarding text, yes: in certain situations the agent can complete transactions and share sensitive information with third parties without first requesting user confirmation.
When will Google officially announce Gemini Spark?
Google I/O 2026 opens May 19 and Gemini is expected to dominate the agenda. No official announcement has been made; the product is currently known only through a leaked onboarding screen inside the Gemini web app.