The adjustment of a small privacy text in Google Chrome caused confusion among users.
Google has updated the description of its on-device AI functionality in its Chrome browser, leading to online concerns about potential changes in how user data is processed. Previously, Chrome stated that AI features ran locally on the device “without sending data to Google servers.” That sentence has disappeared from the settings in the latest update of Chrome 148.
No change in processing according to Google
Due to this adjustment, users thought that Google might start sending interactions with local AI models to the cloud after all. Google denies that anything has changed technically. According to the company, all processing of prompts and responses by Gemini Nano continues to happen locally on the device.
Gemini Nano is the compact AI model that Chrome has been using since 2024 for features such as scam detection and AI-assisted browser functions. The model runs directly on the user’s device, utilizing local storage and processing power.
Confusion caused by new AI features
Google is currently rolling out the new Prompt API. This allows websites to communicate with the AI model running locally in Chrome.
This means that websites can send prompts to Gemini Nano and receive the responses back. This communication happens locally via the browser, but the website itself gains access to the results. The old text could give the impression that data would never go to Google or other websites, whereas a website can, of course, process input and output from the model via the API.
Four gigabytes of local AI
It also doesn’t help that Google automatically downloads its Gemini Nano model on compatible systems. That model takes up approximately four gigabytes of storage space. Google now offers an option to disable and delete the model via the browser settings. Once disabled, the model is no longer downloaded or updated.