Google unveils Gemini Spark at Google I/O 2026
Copyright 2026 The Associated Press. All rights reserved
Key Takeaways
Autonomous Purchasing Warning: Google’s I/O 2026 keynotes emphasized secure, user-authorized checkout protocols, but code in the Google App reveals that Gemini Spark may still make purchases without asking.
Unannounced Usage Caps: Even paying Google One Ultra subscribers will likely face usage caps on Gemini Spark with no obvious mechanism to top up token credits.
Magic Pointer Vs Gemini Spark’s Comet: Two Paths For Google’s AI
Google has announced a raft of new AI tools and agentic features, so it’s easy to become confused: We’ve already seen the new Googlebook’s “Magic Pointer,” activated via a new “Wiggle-To-Wake” gesture that lets the AI see your current window and respond to voice commands like “What does this mean?” or “Move that here.” We’ve also been introduced to Google’s new Gemini Spark agent identity, featuring the familiar Gemini “sparkle” icon flanked by motion streaks symbolizing the agent actively working on tasks in the background. While Magic Pointer serves as an AI helper on your local device, Google’s new agent takes a leap into cloud-based, background execution, bringing with it some serious risks.
Automatic Shopping and Strict Usage Limits
Included in the onboarding text for Gemini Spark is a stark warning that the assistant can make purchases without asking.
Google’s Gemini Spark onboarding warns that it may “make purchases without asking”.
PAUL MONCKTON / GOOGLE
Google Vice President and General Manager of Ads and Commerce, Vidhya Srinivasan, addressed this issue directly in the keynote. She answered the question “How do I know it just won’t go off and buy something I don’t want?” by introducing Google’s new Agent Payment Protocol, AP2, that securely authorizes payments based on pre-authorized user instructions. However, Google still feels the need to protect itself with its catch-all disclaimer, warning that the agent possibly can make unexpected purchases after all.
The onboarding text present in `robin_agent_onboarding_card2_body_spark` includes the following key statements:
“While it is designed to ask for your permission before taking sensitive actions, it may do things like share your info or make purchases without asking. Make sure to supervise Gemini Spark, and don’t rely on it for medical advice, legal, financial, or other professional help.”
So will Spark always check for permission before making a purchase, or will it sometimes just buy stuff without asking? The language suggests that it’s your responsibility to make sure this doesn’t happen.
Usage Limits, Even For Ultra Subscribers
Furthermore, the embedded string `assistant_agent_quota_banner_body_ultra_spark` contains the text:
- “Gemini Spark will be available again when your limit resets.”
The words “ultra_spark” strongly suggest that even Google One Ultra subscribers will be subject to usage limits on Gemini Spark, although it’s not clear whether this applies only to the new lower-cost $100 Ultra subscription or to the newly-reduced $200 Ultra subscription too.
There’s no indication in the code that users will be able to top up this limit by purchasing additional AI credits, unlike other Google AI products like Flow and Antigravity. If this turns out to be true, then Gemini Spark use will maintain its own quotas separate from Google’s unified AI Credits system and means users could potentially face being locked out of Agent use until their limits reset automatically.
Google’s keynote paints a picture of worry-free background automation, as the Google App’s warning text makes clear, it’s still on you, the user, to keep a vigilant eye over Gemini Spark’s background operations and their credit card statements.
I’ll be tracking how Google rolls out these new features as the beta goes live.
