Google Chrome VP and GM Parisa Tabriz has responded to criticism over Chrome’s practice of silently downloading a 4GB AI model onto users’ devices, saying on-device AI is central to the browser’s security and developer strategy.
What triggered the backlash
Privacy researcher Alexander Hanff recently documented the behavior, finding that Chrome automatically downloads Gemini Nano’s model, which is around 4GB in size, on devices without prompting users or offering a clear opt-out. Deleting the file manually triggers an automatic re-download on the next Chrome restart. We recently covered the story and offered instructions on how to turn it off.
The backlash intensified after critics pointed out a confusing inconsistency: Chrome’s highly visible “AI Mode” feature in the address bar did not use the local model at all. Queries typed into it are sent to Google’s cloud servers, meaning users absorb the storage and bandwidth cost of a 4GB file that has no connection to the browser’s most prominent AI feature.
Privacy advocates have also raised concerns about potential violations of EU privacy law, specifically the ePrivacy Directive’s requirement for user consent before storing data on a device.
Google’s response
Tabriz acknowledged the behavior in a recent string of posts on X, clarifying that Google has been offering Gemini Nano in Chrome since 2024 “as a lightweight, on-device model,” that’s core to both Chrome’s developer APIs and its security capabilities, including scam detection.
She noted that the model processes data locally rather than sending it to Google’s servers, and said it automatically uninstalls when a device is low on storage.
Tabriz did not address the consent question directly, nor did she explain why the model reinstalls itself if a user deletes it. Google has said separately that users can disable and remove the model through Chrome’s settings, and that once disabled, it will not re-download.