Even if you’re not using Google Gemini, it may already be on your device. Security researcher Alexander Hanff, known online as “That Privacy Guy,” revealed that Google’s Chrome browser is quietly storing an on-device AI model, which users hadn’t explicitly agreed to.
Hanff reported that Chrome keeps around 4GB of these model files in a folder called OptGuideOnDeviceModel, with a key file named weights.bin. This model appears to be Gemini Nano, Google’s lighter AI version meant for devices with specific hardware capabilities. The unsettling part? Users aren’t prompted to accept its installation, nor are there easy options to remove it.
If you manage to find these files and delete them, they’ll often be redownloaded once you restart Chrome. Hanff tested this on both Windows and MacOS, verifying that the model keeps reappearing without any warning to the user.
When Gizmodo contacted Google for comment, a spokesperson mentioned that Gemini Nano has been available in Chrome since 2024. It helps with important features like scam detection without sending data to the cloud. However, users can easily remove the model if they have the right settings enabled. As of February, Google introduced a feature that allows users to turn off and uninstall the model directly from Chrome’s settings.
Despite reports drawing attention to this issue, many users flagged it on IT forums and social media months ago, unsure about the file’s purpose or how to delete it. Google doesn’t shy away from acknowledging the on-device model exists, but it isn’t exactly highlighted either. Their documentation states that this model supports most AI functions in Chrome, and developers can access it through a specific API.
If you’re concerned about the AI model on your device, first check if it’s there. You can do this by visiting “chrome://on-device-internals” in the Chrome browser, which will show you the status and size of the model.
To remove it, go to the Settings menu in Chrome, and under System, you may find the option to turn the on-device AI on or off. If it’s not there, some users suggest visiting “chrome://flags” to disable AI features and prevent automatic downloads. Another workaround includes setting the weights.bin file to read-only, but keep in mind that this may impact browser performance, especially leading to some AI features not functioning.
The conversation around these AI models reflects larger concerns about user consent and data privacy. According to a recent survey by the Pew Research Center, about 81% of Americans feel they have little to no control over the data collected by companies online. As we increasingly rely on tech, understanding how our data is used—and knowing how to manage it—has never been more critical.
For more insights into how AI impacts our daily lives and the privacy issues surrounding it, check out this Pew Research report.
Source link
ai models,Artificial intelligence,Gemini,Google,Google Chrome,large language model
