Google Chrome has been found to automatically download a 4GB AI model called Gemini Nano onto users’ devices without explicit notification or consent. The discovery was made by security researcher Alexander Hanff, also known as “That Privacy Guy.” This AI model is designed to support features like “Help Me Write,” enabling on-device AI capabilities rather than relying solely on cloud processing. The Gemini Nano model is a lightweight version of Google’s large language models intended for local use on machines that meet specific hardware requirements.
Despite its utility, the silent download of such a substantial file raises privacy and transparency concerns, especially since users are not clearly informed about the download or the data stored on their devices. The model file is stored within Chrome’s installation directory and, if manually deleted, will be redownloaded automatically when Chrome restarts. This behavior has sparked criticism regarding compliance with privacy regulations, particularly under European Union law.
While the on-device model could improve responsiveness and privacy by processing queries locally, users might assume their AI interactions remain entirely on-device when in fact, certain AI functions still route requests to Google’s servers. Google offers settings to opt out of this feature, allowing users to prevent the Gemini Nano model download if desired.
This situation highlights the ongoing tension between deploying advanced AI features for enhanced user experience and maintaining transparency and user control over what data and models are stored locally.
