Tech: How to Disable Google's Gemini in Chrome (2026)
If you use Google's Chrome browser for desktop, there's probably a Gemini Nano AI model running on your computer right now and taking up about 4 GB of space. That's not necessarily a bad thing, but if you didn't know about it and don't want it, there's a way to turn it off. A Google spokesperson tells WIRED that the company started rolling out the On-device AI toggle in February so users can turn off the features if they choose and remove the model. “Once disabled, the model will no longer download or update,” the spokesperson says in a statement. The company added, too, that the system is designed so Gemini Nano “will automatically uninstall if the device is low on resources.” Google built the model into Chrome to enabled on-device AI scam-detection features. It was also aimed at providing a way for developers to integrate AI-related application programming interfaces while keeping data on users' devices when possible and out of the cloud. These features are separate from Chrome's AI Mode, which does not use the local Gemini Nano model. Parisa Tabriz, Chrome's general manager, emphasized in a post on X on Wednesday that integrating Gemini Nano “powers important security capabilities like on-device scam detection and developer APIs without sending your data to the cloud.” Google certainly did announce the Gemini Nano integration into Chrome and discussed it publicly, but for users who simply use Chrome because it is the world's biggest, most recognizable browser and don't necessarily follow every granular update, the lack of an in-your-face notification about a large AI model file sitting and running on your computer may be upsetting. Longtime security and compliance consultant Davi Ottenheimer says that he follows Chrome updates closely but could have easily missed the Gemini Nano integration. “An on-device model could be a hidden minefield,” he says. And the fact that Google launched the integration in 2024 but didn't start rolling out a settings control for
Source: Wired