Gemini Spark's setup screens reveal it may make purchases without asking and impose hidden usage caps, contradicting Google's I/O 2026 keynote messaging.
Google spent considerable time at I/O 2026 reassuring attendees that Gemini Spark, its new autonomous AI agent, would only execute purchases with explicit user authorization. Presenters walked through secure checkout protocols. The message was clear: you stay in control.
The fine print says something different.
Code discovered in the Google App and reported by Forbes reveals that Spark's onboarding flow warns users the agent "may do things like share your info or make purchases without asking." That sentence, buried in setup screens rather than spotlighted in the keynote, shifts the burden of oversight squarely onto users, a notable departure from how Google framed the product publicly.
This is not a minor technical caveat. Gemini Spark is designed to operate in the background, handling tasks autonomously. An agent authorized to initiate financial transactions without prompting is a materially different product from one that always checks first. Google has not publicly addressed the discrepancy between its keynote claims and the language surfaced in the app code.
A second omission compounds the first. Even subscribers paying for Google One Ultra, the company's top-tier plan, will apparently face usage caps on Gemini Spark, with no visible mechanism to purchase additional credits once those limits are hit. Users who reach the ceiling could find the cloud-based agent simply unavailable until quotas reset, a hard stop that sits oddly against Google's vision of seamless, always-on automation.
Both gaps raise a pointed question about how Google defines agentic AI in practice versus in marketing. Autonomous artificial intelligence agents that interact with financial accounts require a higher standard of disclosure than agents that merely answer questions. The distance between keynote polish and onboarding text suggests the product's governance model was still developing at launch.
The competitive stakes
Google is not alone in navigating these tensions. The broader industry is under growing pressure to find sustainable monetization beyond venture capital. Digital Watch Observatory tracked how OpenAI's January 2026 launch of ChatGPT Go, an $8-per-month tier, marked a turning point in how AI companies approach paid access while resisting advertising. Every major player is working through the same calculation: extract revenue without eroding the trust that makes the product worth paying for.
Agentic features are central to that bet. An AI that can autonomously book travel, order groceries, or renew subscriptions becomes genuinely indispensable, and indispensability justifies subscription revenue. But that value proposition depends entirely on users trusting the agent to stay within its mandate. A single unwanted charge can undo months of goodwill, and Forbes recommends users actively monitor all financial activity Spark initiates.
Regulators are also likely to take notice. EU and UK authorities have shown growing appetite for scrutiny of artificial intelligence systems that interact with payment data or execute transactions on users' behalf. Burying authorization scope in onboarding fine print rather than surfacing it in public disclosures is precisely the pattern that draws regulatory attention.
What users should do now
Until Google clarifies how Spark's authorization model actually works, anyone enabling the agent should assume it has broader financial permissions than the keynote suggested. Review linked payment methods before activation, enable transaction alerts, and read the full setup text carefully. The Forbes report makes clear that passively trusting the product carries real financial risk.
Monitoring the usage cap situation matters too. Paying subscribers who hit Spark's limits with no obvious path to more credits will find the product substantially weaker than advertised, especially if the agent goes dark mid-task. That is a product management failure as much as a transparency gap.
Whether Google's disclosure practices for agentic artificial intelligence are adequate for the responsibilities these tools carry is now a live question. Spark is not a chatbot; it is software with the authority to act in the world on a user's behalf. That requires a different level of clarity than a keynote can provide.
Google has not commented on the discrepancy as of publication time.
---
FAQ
Can Gemini Spark make purchases without my permission?
Based on onboarding text found in the Google App, yes. Setup screens warn the agent "may do things like share your info or make purchases without asking," a disclosure that contradicts the user-authorized checkout model Google described publicly at I/O 2026.
Does Google One Ultra include unlimited Gemini Spark usage?
No. Even top-tier subscribers are expected to face usage caps on Spark, with no apparent option to buy additional credits once limits are reached.
What is Gemini Spark?
Google's new autonomous AI agent, announced at I/O 2026, designed to handle tasks in the background including, potentially, financial ones on the user's behalf.
How do I protect myself when using Gemini Spark?
Review linked payment methods before enabling the agent, activate transaction notifications, and read all onboarding text in full. Treat Spark as having broader permissions than Google's announcements imply until the company provides clearer documentation.