Google may be preparing to take a major step beyond traditional AI assistants with a new experimental system reportedly called Gemini Spark. Ahead of next week’s Google I/O 2026 event, leaked onboarding screenshots have started circulating online, suggesting the company is working on a far more autonomous version of Gemini — one that could actively perform tasks across apps, websites, and connected services instead of simply responding to prompts.
If the leak turns out accurate, Gemini Spark could represent one of Google’s biggest AI experiments so far. Unlike regular chatbot-style assistants that wait for users to type commands, Spark is reportedly designed to operate continuously in the background and proactively help users complete tasks. That shift may sound convenient on paper, but it also raises serious questions about privacy, permissions, and how much control users are willing to hand over to AI systems.
According to the screenshots shared online, Gemini Spark may gain access to a huge amount of user information and connected services. The leaked onboarding process reportedly mentions access to connected apps, chats, tasks, websites users are signed into, personal intelligence features, locations, and even browser-related data. In simple terms, this sounds less like a chatbot and more like an always-running AI layer deeply integrated into a person’s digital life.
The feature is reportedly still experimental, and Google appears aware of the risks involved. The onboarding warnings allegedly state that Gemini Spark may ask for permission before carrying out sensitive actions, but there could also be situations where it performs tasks like sharing information or making purchases automatically. That part alone will likely trigger concern among many users because it moves AI assistants closer to acting independently rather than simply offering suggestions.
One of the more surprising details in the leak is the mention of remote browser data storage. Reports suggest Gemini Spark may save login credentials and other browser-related information to carry out actions across websites. The leaked details also referenced remote code execution capabilities, which sounds extremely powerful but also potentially dangerous if misused or exploited improperly.
To be fair, Google reportedly plans to offer controls allowing users to clear stored data and disable Connected Apps or Personal Intelligence features through settings. Activity management options tied to Gemini Apps Activity controls may also give users ways to monitor or delete AI-related actions later. Still, many privacy advocates will probably question whether average users fully understand what kind of access they are actually granting these AI systems.
The timing of the leak is not surprising because the AI industry is currently moving aggressively toward “AI agents” rather than simple chatbots. Companies across Silicon Valley are racing to build systems capable of performing multi-step tasks independently. OpenAI, Anthropic, and even Microsoft have all been experimenting with AI agents that can browse websites, use software tools, and automate workflows with minimal user involvement.
Google clearly does not want to fall behind in that race. The company has already spent the last year embedding Gemini across Android, Workspace, Search, and Chrome. But Spark appears to take that idea much further by transforming Gemini into something closer to a digital operator managing actions on behalf of users.
At the same time, this kind of technology also enters risky territory very quickly. An AI system that can access browser sessions, connected accounts, location data, and financial information creates enormous convenience potential — but also enormous security concerns. Even Google’s own onboarding reportedly warns users not to rely on Spark for medical, legal, or financial advice, which suggests the company still sees major limitations and risks around autonomous AI behavior.
The leak also comes during a period when regulators globally are paying closer attention to AI systems that collect and process personal data. Europe especially has become more aggressive regarding digital privacy laws, and features like Gemini Spark could easily face scrutiny if they appear too invasive or difficult for users to control transparently.
Still, from a technological perspective, Gemini Spark may preview where consumer AI is ultimately heading. Instead of users constantly opening apps, searching manually, and moving between services, future AI systems may quietly handle many of those workflows automatically behind the scenes. Booking tasks, organizing information, managing schedules, purchasing products, and coordinating apps could eventually happen through autonomous agents rather than direct user input.
Google is expected to reveal major AI announcements during Google I/O 2026 beginning May 19, so it would not be surprising if Gemini Spark becomes one of the event’s biggest talking points. But if the company officially unveils it, the conversation likely won’t only be about innovation. It will also be about how much trust users are actually willing to place in AI systems that operate constantly in the background of their digital lives.
