New Delhi. An experimental Google AI assistant named COSMO briefly appeared on the Google Play Store ahead of Google I/O 2026 before being removed, triggering wide discussion in the tech community. The 1.13 GB test application was reportedly not intended for public release, but users who downloaded it during the short window found features suggesting a more proactive form of AI assistance.
Assistant Designed to Work Without Direct Commands
COSMO is being described as different from conventional AI assistants because it does not simply respond to direct user commands. According to leaked details, the system can analyse user behaviour, screen activity and digital patterns to understand when assistance may be needed.
The assistant reportedly functions as an intelligent layer within a smartphone, reading the device screen in real time and interpreting ongoing activity. If a user is planning a trip in a chat, it may create a calendar event. If a user is researching a topic, it may gather information from multiple sources and present a structured summary.
Screen Awareness and Automation Draw Attention
One of the key features attributed to COSMO is real-time screen awareness. Reports indicate that it uses Android system-level tools to understand on-screen content and context.
The system is also said to go beyond making suggestions. It may be able to perform browser automation tasks, including opening websites, searching for information and carrying out certain actions independently. This positions COSMO as a more automated digital assistant rather than a standard chatbot.
Gemini Nano Link Raises Privacy Questions
COSMO is reportedly built on Google’s Gemini Nano model, allowing it to run directly on the device and function even without an internet connection. For more complex tasks, it is said to connect with cloud-based systems, creating a hybrid processing model aimed at balancing performance and privacy.
It remains unclear whether Google plans to launch COSMO as a separate product or integrate it into the wider Gemini ecosystem. While the leak has generated interest in the future of AI assistants, it has also raised concerns about privacy, user control and continuous screen monitoring.
The accidental exposure has offered a glimpse into how AI assistants may evolve from passive tools into active digital systems capable of understanding context and completing tasks with limited manual input.
