Google caught automatically downloading 4GB AI model on Chrome without consent

Google Chrome has been silently downloading a 4GB AI model to devices

Alexander Haniff, the AI researcher known as “That Privacy Guy“, discovered this week that Google has been silently downloading a 4GB AI model onto the systems of Google Chrome users. This model was downloaded without users’ consent and redownloads itself when users delete it.

Right now, Google Chrome is the world’s dominant web browser. Google Chrome is the most used web browser, accounting for 67.97% of web users. That’s over three billion people. Not only has Google wasted countless terabytes of storage by adding this AI model to systems, but it has also forced a feature that many users would actively want to avoid.

For PC users with an internet usage cap (data cap), Google has wasted 4GB of your limited internet. Add the environmental impact of pushing a huge 4GB update onto billions of users who didn’t request it, and it’s easy to see why many people are annoyed with Google over this. They have violated user trust, wasted resources, and given users no opt-out. Thanks, Google…

This week I discovered the same pattern, executed by Google. Google Chrome is reaching into users’ machines and writing a 4 GB on-device AI model file to disk without asking. The file is named weights.bin. It lives in OptGuideOnDeviceModel. It is the weights for Gemini Nano, Google’s on-device LLM. Chrome did not ask. Chrome does not surface it. If the user deletes it, Chrome re-downloads it.

– Alexander Haniff – “That Privacy Guy”

What’s this secret Google Chrome AI model download for?

Google claims that its AI model download for Chrome is for “Gemini Nano”, a lightweight on-device model that can power certain AI capabilities on-browser. Google claims that this enables features like scam detection to run in your browser without sending data to the cloud.

Regarding the model’s download size, Google has claimed that it automatically uninstalls itself if a device is low on resources. Furthermore, they claim that they have started rolling out the ability for users to “easily turn off and remove the model”.

We’ve offered Gemini Nano for Chrome since 2024 as a lightweight, on-device model. It powers important security capabilities like scam detection and developer APIs without sending your data to the cloud. While this requires some local space on the desktop to run, the model will automatically uninstall if the device is low on resources. In February, we began rolling out the ability for users to easily turn off and remove the model directly in Chrome settings. Once disabled, the model will no longer download or update. More details in our help center article.

– Google to Android Authority

Google, this is not good enough

While rolling out the ability to “turn off and remove” this AI model is good, it shouldn’t have been delivered without users’ consent in the first place. Users won’t know to uninstall something they don’t know exists. Furthermore, given the size of the download for a non-essential feature, it should have been optional.

Because this model automatically deletes itself on systems with low resources, some users with a borderline level of available storage may see it frequently downloading and later deleting itself. That’s a lot of wasted data.

Google could have easily provided users with a prompt that explained their Gemini Nano model. They could have told them about their on-browser AI security features and their benefits. Users could have had the option to download the AI model or not. If Google had given users a single prompt, its current bad press could have been avoided. Instead, Google has broken user trust and wasted countless resources on features that many would delete at the first opportunity. Well done Google.

You can join the discussion on Google’s automatic Gemini Nano download on the OC3D Forums.