Chrome's AI Model Deployment Raises Privacy and Transparency Concerns

Google's Chrome browser downloads a 4GB AI model for on-device processing, a practice initiated in 2024. This model supports features like Gemini chatbot integration and web browsing automation.

Recent user observations of increased storage consumption have led to confusion regarding the model's deployment. Google states the model's installation depends on hardware, account features, and website interactions.

While local AI processing offers privacy benefits by keeping data on-device, a recent change in Chrome's settings toggle removed the explicit assurance that the on-device AI model would not send data to Google's servers. Google clarified this change was for clarity regarding API functionality, asserting that data processed by the model remains on the device.

However, if a website utilizes Chrome's on-device AI API, the website itself can access input and output data. This means data could reach Google's servers if a Google-owned website uses the API, or remain private if a non-Google site is involved.

This situation highlights ongoing concerns about user control and transparency in AI feature deployment, particularly as AI integration becomes more pervasive across Google's products.