Chrome users discover hidden 4GB Gemini AI file on devices — here’s what we know

Google Chrome is quietly downloading a massive artificial intelligence model directly onto users’ devices, raising serious privacy and storage concerns among cybersecurity experts.

According to a recent investigation by Cybernews and privacy researcher Alexander Hanff, the popular web browser is silently installing a 4GB file named “weights.bin” without explicit user consent.

This hefty file is hidden within Chrome’s system folders, specifically in the OptGuideOnDeviceModel directory, and runs entirely in the background without any notification or opt-out interface.

The massive download is tied to Gemini Nano, Google’s lightweight on-device large language model. This AI engine is designed to power new browser features natively, such as intelligent writing assistance, rapid text summarization, and real-time scam detection.

By processing this sensitive data directly on the computer rather than routing it through cloud servers, Google aims to enhance user privacy and significantly improve offline functionality.

However, the storage tradeoff has left many users frustrated. Because Gemini Nano requires massive sets of trained parameters to function, the weights.bin file instantly consumes roughly 4GB of precious disk space.

For users operating laptops with limited storage, this sudden, unannounced space reduction poses a severe technical problem.

Furthermore, researchers have found that Chrome does not consistently remove previous model versions when downloading updates. This oversight has led some users to find multiple version directories hoarding 12GB or more of their local storage.

While users can manually locate and delete the file from their Windows or Mac system folders, Chrome is programmed to automatically re-download it if the related AI features remain active.

To permanently halt the secret installation, users must navigate to the browser’s hidden developer settings at “chrome://flags” and manually turn off specific experimental features, such as the OnDeviceModelBackgroundDownload flag.

While the industry shift toward on-device AI offers undeniable privacy and speed benefits, the sheer lack of transparency surrounding this massive background download has generated intense backlash.

As artificial intelligence becomes deeply embedded in everyday digital tools, the debate over explicit user consent and local storage limits is rapidly intensifying across the tech landscape.