Please follow us on Gab, Minds, Telegram, Rumble, Truth Social, Gettr, Twitter, Youtube  

Google Chrome is automatically downloading a roughly 4-gigabyte on-device AI model to users’ computers, often without their knowledge or explicit permission, according to privacy researchers and technical analyses.

The model in question is Gemini Nano, Google’s lightweight version of its AI system designed to run locally on user devices. It powers features such as “Help me write” text generation, on-device scam detection, and a Summarizer API that allows websites to access its capabilities directly.

Privacy and security blogger Alexander Hanff first drew widespread attention to the practice in a May 4 blog post. The AI model is stored as a file named weights.bin inside a directory called OptGuideOnDeviceModel within Chrome’s user profile folder.

According to Hanff and other researchers, the model downloads automatically when AI-related features are enabled in Chrome — features that are turned on by default in recent versions of the browser. Even if users manually delete the file, Chrome often redownloads it unless specific experimental flags are disabled.

Reports suggest the practice has been underway for some time. A Reddit discussion from April 2025 referenced a roughly 3 GB version of the model, while a November 2025 Stack Overflow thread indicated it had grown to approximately 4 GB.

Environmental and Cost Concerns Raised

Critics have highlighted the significant resource implications of silently distributing such a large model to Chrome’s massive user base. If deployed to just one billion users — roughly 30 percent of Chrome’s total install base — the initial downloads alone could consume an estimated 240 gigawatt-hours of electricity and generate around 60,000 tons of carbon dioxide equivalent, researchers estimate. Those figures do not include ongoing energy use during model operation.

Bandwidth costs are another point of contention. While the download may seem minor for users with high-speed, unlimited connections, it can quickly exhaust data caps for those on metered plans, mobile hotspots, or rural internet services.

“For rural users or those with bandwidth caps, this kind of silent transfer can blow through monthly limits in minutes,” a Malwarebytes report noted.

Privacy Questions Emerge

Privacy advocates have also expressed concern over the deployment method and what it signals about user control.“

All in all, users see a 4GB local AI model and reasonably assume their data stays private, when in reality, the most visible AI feature sends everything to Google’s servers,” Malwarebytes stated. “Tech companies need to stop treating silent deployment as acceptable practice. We see no valid excuse for this. Your device is yours. The storage is yours. The bandwidth is yours. And the electricity bill is yours.”

To prevent the model from downloading, users can disable the relevant Chrome flag. Type chrome://flags into the address bar, search for “Optimization Guide On-Device,” and set the dropdown option to Disabled. Restarting the browser should stop the background installation.

The discovery has sparked renewed debate over how much control technology companies should exercise over users’ hardware and data usage, particularly as AI features become more deeply integrated into everyday software.

‘NO AD’ subscription for CDM!  Sign up here and support real investigative journalism and help save the republic!