Chrome automatically downloads a 4 GB AI model without user consent

Chrome automatically downloads a 4 GB AI model without user consent

Chrome downloads a local AI model of 4 GB to every computer without asking. Google does not request permission and does not show a notification — the browser simply does it as a fact.

What's happening in Chrome

Chrome automatically loads the weights of the Gemini Nano model into the OptGuideOnDeviceModel folder. This is a weights.bin file that contains the weights of Google's local language model. The size is 4 gigabytes. The browser downloads it without any consent windows, without notifying the user. If a user manually discovers and deletes the file, Chrome simply downloads it again on the next launch. It is unclear how exactly Google plans to use this model on the device. Possibly for text analysis directly in the browser, without sending to servers. But the fact that the file is downloaded without consent already violates the user's expectations about what their browser can do.

History repeats itself

Two weeks ago, a similar scandal erupted around Claude Desktop from Anthropic. The application registered a Native Messaging module in seven Chromium-based browsers at once. This allowed Claude to inject itself into the settings of other programs without user consent. The scheme was simple: installed Claude Desktop — and it automatically "rewrote" itself into Chrome, Edge, Brave and other browsers. Neither a consent window nor a visible way to refuse in the Claude interface itself. If a user manually deleted the entry — it would appear again on the next launch. Now Google is using a similar approach, but much more invasive:

• The file is downloaded without asking the user
• There are no notifications in the browser interface
• There is no visible way to prevent this
• The file is restored automatically when deleted
• It affects billions of devices around the world

The scale of environmental harm

Here Google differs from Anthropic precisely in scale. Chrome is installed on approximately 3.5 billion devices worldwide. When a company decides to roll out 4 gigabytes of binary files to this entire audience, the environmental cost becomes truly material. According to calculations, one wave of such downloads costs approximately 6-60 thousand tons of CO2 equivalent. Server bandwidth, energy on servers, energy on user devices — it all adds up. For perspective: an average car produces approximately 4-5 tons of CO2 per year. That is, Google's decision can generate as many emissions as 1.5 to 15 million cars running for a whole year.

The problem of trust

Behind these two incidents lies a deeper problem: the boundaries of trust between vendors are becoming increasingly blurred. Companies are taking the right to download and install something on other people's devices without explicit consent. They justify this with feature improvements, the need to work locally, increased privacy. But in reality, a situation is created where the user loses control over their own device. You cannot prevent Chrome from downloading files, you cannot see what exactly is being downloaded, you cannot disable it in the interface. The file simply appears on your disk, takes up space and electricity.

What does this mean

History shows an old problem in a new guise: large vendors are taking the right to make decisions for their users. Privacy, environmental damage, device control — all of this is left aside. When there are hundreds of such decisions, and each one affects billions of people, the problem takes on a global scale.