Gemini Spark Raises Concerns Over Autonomous Purchases and Usage Caps

Google's latest AI agent, Gemini Spark, unveiled at the recent I/O 2026 conference, has quickly become a subject of concern due to revelations about its potential for autonomous actions and unstated usage restrictions. While Google's keynote presentations highlighted robust, user-authorized checkout protocols for Spark, internal code within the Google App indicates a more ambiguous reality. This code suggests that Gemini Spark might proceed with purchases without explicit user consent, with onboarding text reportedly warning that it "may do things like share your info or make purchases without asking," effectively placing the onus of supervision on the user. This potential for unsupervised transactions directly conflicts with the assurances provided by Google's Vice President and General Manager of Ads and Commerce, Vidhya Srinivasan, who introduced the Agent Payment Protocol (AP2) as a secure method for authorizing payments based on pre-approved user instructions. The discrepancy between public statements and internal code raises significant questions about the level of control users will truly have over Gemini Spark's financial actions. Adding to these concerns are indications of unannounced usage caps for Gemini Spark. Even subscribers to the high-tier Google One Ultra plan are expected to face these limits. The internal code, specifically mentioning "ultra_spark," suggests that these caps will apply to Ultra subscribers, though it's unclear if this extends to both the new lower-cost and the newly-reduced higher-tier Ultra subscriptions. Crucially, there appears to be no mechanism within the code for users to purchase additional AI credits to bypass these limits, unlike other Google AI products. This lack of transparency around usage limits, coupled with the potential for unauthorized purchases, presents a challenging outlook for user trust and the practical deployment of Gemini Spark.