Google accidentally reveals COSMO AI assistant app ahead of I O 2026 rollout hinting at next generation on device intelligence

The COSMO application was listed on the Play Store on May 1 under the package name com.google.research.air.cosmo, clearly indicating its origin within Google Research. Unlike typical experimental releases, however, the app was not distributed through a limited beta or labs channel but instead appeared on Google’s primary Play Store account, making it briefly accessible to a wider audience.

Described as “an experimental AI assistant application for Android devices,” the listing suggested that COSMO is not merely a concept project but a functional testbed for next generation AI capabilities. The app was quickly taken down, reinforcing the likelihood that the release was accidental rather than part of a planned rollout.

Despite its short availability window, enough details were captured to build a clear picture of its purpose and design.

One of the most notable aspects of COSMO is its integration of a full local AI model. The app, which measured approximately 1.13 GB in size, included the on device version of Google’s AI system, .

This is significant because it reflects a broader industry shift toward running AI directly on devices rather than relying entirely on cloud processing. On device models offer several advantages, including faster response times, improved privacy, and reduced dependence on internet connectivity.

COSMO appears to take this concept further by offering multiple inference modes within its settings. Users could choose between:

This flexible architecture suggests that Google is experimenting with balancing performance, privacy, and computational efficiency in real world scenarios.

Beyond its underlying AI model, COSMO stands out for its extensive list of built in features referred to as “Skills.” The app reportedly included 14 such capabilities, though not all were enabled by default during testing.

These Skills indicate a clear attempt to move beyond traditional voice assistants toward a more proactive and task oriented AI system. Among the most notable features were:

Together, these features suggest that COSMO is being positioned as more than just a conversational assistant. Instead, it appears to be evolving into a comprehensive digital companion capable of handling complex workflows.

Another key detail uncovered during testing is COSMO’s use of the AccessibilityService API within . This system level access allows applications to read and interact with on screen content, enabling deeper integration with device activities.

While this opens the door to powerful capabilities such as screen understanding and contextual assistance, it also raises important questions around privacy and security. According to early observations, this functionality was not fully operational in the leaked version, suggesting it is still under development.

Google has historically taken a cautious approach to sensitive permissions, and it is likely that any final implementation will include strict safeguards and user controls.

Several elements of the Play Store listing point toward an unintended release. Screenshots were reportedly displayed in incorrect aspect ratios, and compatibility appeared inconsistent even on Google’s own Pixel devices.

Such issues are uncommon in official launches, especially for a company with Google’s level of polish and quality control. These inconsistencies reinforce the conclusion that COSMO was not ready for public exposure and was likely published prematurely.

The swift removal of the app further supports this interpretation.

One of the biggest unanswered questions is how COSMO fits into Google’s existing AI ecosystem, particularly alongside its Gemini app and services.

There are several possibilities:

Given Google’s recent emphasis on unifying its AI offerings under the Gemini brand, the second scenario appears especially plausible. COSMO may function as a proving ground where new ideas are refined before reaching a broader audience.

With just weeks away, industry observers expect the company to address COSMO directly or indirectly during its keynote presentations.

Google I O has traditionally been the stage for major announcements related to Android, AI, and developer tools. If COSMO represents a significant step forward in on device intelligence, it is likely to play a role in the company’s broader narrative around the future of computing.

The timing of the accidental release strongly suggests that COSMO or its underlying technologies were intended to be unveiled at the event.

The emergence of COSMO comes at a time when competition in the AI assistant space is intensifying. Companies across the technology sector are investing heavily in generative AI, automation, and personalized digital experiences.

Google’s approach, as reflected in COSMO, appears to emphasize:

If successfully implemented, these elements could position Google as a leader in the next generation of AI assistants.

While the premature appearance of COSMO on the Play Store may have been unintended, it has provided a rare and valuable look into Google’s ongoing AI development efforts.

The app reflects a clear shift toward more intelligent, capable, and autonomous assistants that operate seamlessly across devices and contexts. By combining on device processing with advanced features and flexible architecture, COSMO hints at a future where AI becomes deeply embedded in everyday digital experiences.

All eyes will now be on Google I O 2026, where the company is expected to clarify its vision and potentially introduce COSMO or its underlying technologies to the world in a more formal and complete manner.