Google Chrome Secretly Downloads 4GB AI Model, Sparks Privacy Backlash

Google Chrome has quietly begun deploying Google’s Gemini Nano, a 4GB artificial intelligence model, onto user devices without explicit consent. This unannounced download, stored as weights.bin, integrates the foundational model directly into the browser’s file system, enabling local AI execution via the newly introduced Prompt API for JavaScript. Affecting potentially over a billion users, this translates to billions of gigabytes of unrequested data, with Chrome automatically redownloading the model if users attempt to delete it. While the Prompt API promises powerful on-device AI capabilities—such as local conversational agents, multimodal processing, and integrated developer tools without server-side interaction—the deployment method has ignited significant backlash. Users can verify the model’s presence and status via chrome://device-internals and, if desired, disable it through chrome://flags by deactivating “Prompt API for Gemini Nano” and “Optimization Guide on Device” before manually deleting the file.

The unilateral rollout has drawn criticism from the wider web development community and competing browser projects. Mozilla, Webkit, Microsoft, and the W3C TAG have voiced strong concerns regarding the standardization and implementation of the Prompt API, highlighting a perceived disregard for collaborative web standards given Chrome’s dominant market share. This action follows a broader trend of Google exerting greater control over web access; a recent “Cloud Trust Defense” initiative, an evolution of reCAPTCHA, has raised additional alarms by implicitly requiring Google Play Services for accessing certain websites. This effectively penalizes users of privacy-focused Android distributions like GrapheneOS and CalyxOS by treating their systems as “suspicious,” further underscoring Google’s increasing leverage over web autonomy.