Technology

Google Accidentally Leaked COSMO — Its Most Ambitious (and Terrifying) AI Assistant Yet

|Author: Viacheslav Vasipenok|3 min read| 118
Google Accidentally Leaked COSMO — Its Most Ambitious (and Terrifying) AI Assistant Yet

For a few hours on May 1, 2026, a mysterious app called COSMO appeared on the Google Play Store before being swiftly removed. But the internet never forgets.

What leaked is not just another chatbot. COSMO appears to be Google’s all-in-one, deeply personal AI super-assistant — a genuine attempt to create an always-on intelligence that knows everything about your life.

What We Know So Far

Google Accidentally Leaked COSMO — Its Most Ambitious (and Terrifying) AI Assistant YetAccording to screenshots and descriptions circulating online (detailed by 9to5Google), COSMO packs an impressive set of capabilities:

  • Recall — A direct answer to Microsoft’s controversial Recall feature. The AI continuously remembers everything you do on your device and can bring up context later.
  • Conversation Summary — It listens to your voice calls and reads your messaging apps, then intelligently summarizes conversations and creates calendar events. (“You and Vadim agreed to drink beer behind the garage on Friday — I’ve already added it to your calendar.”)
  • Mariner — An autonomous agent that can browse the web and perform complex tasks on your behalf.
  • Deep Research — Capable of conducting thorough research across sources and delivering structured reports.

The app is unusually large (over 1 GB) because it includes Gemini Nano running locally on-device, allowing many features to work offline. Users can reportedly switch between local, cloud, and hybrid modes.

To function at full power, COSMO naturally asks for virtually every permission possible — which makes perfect sense for an assistant designed to be deeply embedded in your daily life.

Google I/O Warm-Up?

The timing is hardly coincidental.

Google Accidentally Leaked COSMO — Its Most Ambitious (and Terrifying) AI Assistant YetWith Google I/O just days away, this looks like either:

  1. A premature leak by an overeager team, or;
  2. A deliberate teaser to build hype.

Either way, it shows Google is going extremely hard on agentic AI — systems that don’t just answer questions but actively understand your life, anticipate needs, and act on your behalf.

Also read:

The Privacy Elephant in the Room

Google Accidentally Leaked COSMO — Its Most Ambitious (and Terrifying) AI Assistant YetWhile the features sound incredibly useful, they also represent one of the most privacy-invasive AI products ever conceived by a major tech company. A local model helps with some concerns, but the combination of Recall, conversation monitoring, and deep system access is bound to spark intense debate.

Microsoft already faced massive backlash for its Recall feature. Google is walking straight into the same minefield — except with potentially even deeper integration.

We’ll likely get the full official story (and marketing spin) very soon at Google I/O. Until then, COSMO serves as a fascinating preview of what Google believes the future of personal AI should look like: less like a helpful chatbot, and more like a digital second brain that never forgets anything.

Welcome to the age of the all-seeing, all-remembering AI assistant.

Whether that’s exciting or dystopian… well, that’s for each user to decide.

Share:
0