Google Chrome reportedly downloads a 4 GB AI model onto PCs without user consent, raising alarms over transparency. The Gemini Nano model, used for on-device tasks like scam detection, installs silently via an OptGuideOnDeviceModel folder containing a weights.bin file. Discovered by privacy researcher Alexander Hanff, this process occurs automatically when AI features activate, leaving no opt-out toggle for average users.

The 4 GB binary download persists even after manual deletion, forcing users to disable Chrome’s AI tools via chrome://flags settings, enterprise policies, or uninstallation. Google’s lack of prior notification or granular consent options appears to violate GDPR’s Article 5(3) and the UK’s Data Protection Act, which mandate explicit user consent before data access. Similar rules under California’s Consumer Privacy Act further complicate compliance, suggesting potential legal risks for the tech giant.

The absence of visible indicators or warnings during installation underscores a design flaw that prioritizes automation over user agency. While Google claims the model improves local AI performance, critics argue the covert approach undermines trust. For now, power users must navigate obscure settings to halt the downloads, leaving most unaware of the 4 GB AI footprint on their devices.

This user consent breach highlights the need for clearer disclosure practices in AI-driven software updates, particularly as on-device models grow in size and scope.