Google tweaks Chrome "On-device AI" privacy wording; says processing remains on-device

Google has updated the wording in Chrome’s System settings that describes its on-device AI features, removing the phrase "without sending your data to Google servers." The edit prompted questions about whether the company had changed how it handles local AI processing.

Google says the change does not alter the architecture: data passed to the on-device model is still processed solely on the user’s device, the company told reporters.

The company said the wording was adjusted to avoid misinterpretation when websites — including Google sites — interact with the on-device Gemini Nano model via Chrome’s Prompt API. In those cases, the website (not Google’s servers) can access the model’s inputs and outputs, and the resulting data handling falls under the website’s privacy policy.

Separately, Chrome has been downloading the 4GB Gemini Nano model onto users’ devices. That download is not opt-in by default, though Google introduced a way earlier this year for users to deactivate and remove the model from their devices.

The wording change and the rollout of the Prompt API happening at the same time prompted privacy questions because of the possibility that web pages could see prompts and responses from a browser-resident model. Google says the edit was intended to reduce confusion about those interactions and how data is handled when a website calls a user’s on-device model.