Google Tweaks Chrome AI Privacy Wording Over Prompt API Use
Google Tweaks Chrome AI Privacy Wording Over Prompt API Use
https://winbuzzer.com/2026/05/11/google-tweaks-chrome-ai-privacy-wording-insists-pr-xcxwbn/
Publish Date: 2026-05-11 07:00:00
Source Domain: winbuzzer.com
TL;DR
- Settings Edit: Google has removed Chrome wording that said on-device AI would not send user data to its servers.
- Model Downloads: Chrome can download Gemini Nano in the background or on demand before AI features become available.
- Prompt API: Websites may invoke the browser model and may see Gemini Nano inputs and outputs during interactions.
- User Control: Chrome says disabling On-device AI lets users remove Gemini Nano and stop later downloads or updates.
Google appears to be narrowing how Chrome describes its local AI behavior in the app’s settings. Google changed Chrome’s System settings wording, removing a sentence that said on-device AI could run without sending user data to Google servers. With the edit, users have lost the clearest plain-language privacy promise attached to Chrome’s local model controls.
Chrome still uses on-device AI models. Chrome was already under scrutiny before that wording change. Users had been watching model downloads, disk usage, and browser-side AI controls more closely for months.
Chrome had already integrated Gemini Nano into Chrome desktop in 2024. In 2025, Chrome was already using Gemini Nano for on-device scam detection. On January 20, Chrome added a control to turn off on-device AI models.
Google is not answering a claim that Chrome just moved local inference into the cloud. A different question now hangs over the settings text: why did Chrome drop its strongest privacy sentence after AI models were already being installed, removed, and called by more than one feature path?
Why the Prompt API Raised the Stakes
Chrome’s websites can call Gemini Nano inside the browser moment changed the setting’s audience. Earlier wording mainly addressed a user deciding whether Chrome’s own AI features felt safe to leave enabled. A website-facing API widened that frame. Once page code can invoke a browser-resident model, the settings text has to explain how the model reaches the device, what waits for it, and what a…