PSAs regarding app security have been around for ages – outlets, media, and even official bodies are telling people to know what they’re installing and using, basically raising the awareness of the population within the digital world.
What happens if something you trusted thinks that it can circle around your consent and installs something that is “best for you”? That’s what Google is doing with Chrome and Gemini Nano.
Recently, it is known that Google has been pushing a local version of its Gemini Nano directly into Google Chrome, and that’s fine for its intended purpose, since cloud-based Gemini models can be accessed online via Android devices as well as through the URL bar, but in Chrome’s case, that’s local, and they are putting a 4GB model inside your system without telling you.
It is mostly tied to AI-driven features like writing assistance, smarter autofill, and even security-related functions such as scam detection. On paper, that sounds like a forward-looking move, especially since running AI locally can reduce reliance on cloud processing and is usable even if the device is being taken offline.
But users aren’t being clearly told that a multi-gigabyte component is being downloaded onto their systems, nor are they being given a straightforward, upfront choice to accept or decline it. Instead, it’s happening quietly in the background, bundled into what looks like a routine browser update.
And get this – 4GB isn’t that big of a deal in terms of storage on a relative scale, but if loaded into RAM, it can straight up take a big chunk of memory away from the user, throttling overall performance if the device is already doing something else. Factoring that most systems are running only 16GBs of RAM on top of Chrome tabs consuming them like no tomorrow, it can cause a real headache down the line.
There’s also a broader concern here about trust. Even if the intention behind on-device AI is to improve privacy by keeping certain processes local, that benefit is undermined if the rollout itself lacks transparency. Some reports even suggest that certain interactions may still involve server-side processing, which further complicates the narrative around “privacy-first” AI.
All in all, it’s the good ol’ way of saying “Good intention doesn’t always lead to good execution”. I’m all for local AI processing, but it has to be done in a way that respects people’s choice in the long run.
How to check
Find the “3 dots” on the top right, head into Settings, then the System tab on the left, and if “On-Device AI” is being displayed, make sure that it is turned off.
Alternatively, head over to C:\Users[YOUR USER NAME]\AppData\Local\Google\Chrome\User Data, locate the “OptGuideOnDeviceModel” folder, and a file named “weights.bin” should be inside. Feel free to delete it.
The Registry Way
- Search for “regedit” in Start Menu
- Navigate to HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Google\Chrome
- Right-click on the Chrome folder, choose “New”, then “D-WORD (32-bit Value)”
- Name the entry “GenAILocalFoundationalModelSettings” (remove the quotations by the way)
- Put 1 as the value
For users with Administrator privileges who can execute Terminal commands, just paste the following and run:
reg add “HKLM\SOFTWARE\Policies\Google\Chrome” /v “GenAILocalFoundationalModelSettings” /t REG_DWORD /d 1 /f
