Google Chrome has been quietly installing Gemini Nano, a 4GB AI model, onto users' devices without explicit consent. The on-device model powers Chrome's new AI features - writing assistance, tab organization, image generation. But the size and automatic deployment has users asking a reasonable question: shouldn't we get to decide if we want gigabytes of AI models eating our storage?

On-device AI is genuinely the future. Running models locally means faster responses, better privacy, offline capability. From a technical perspective, Gemini Nano is impressive - a capable language model compressed enough to run on consumer hardware without destroying battery life or requiring constant internet connectivity.

The problem isn't the technology. The problem is Google deciding that everyone using Chrome wants 4GB of their storage allocated to an AI model they never asked for. That's larger than many applications. That's comparable to a full game install. And it happened automatically, without a clear prompt asking for permission.

Here's how it works technically. Chrome downloads Gemini Nano in the background as part of regular browser updates. The model gets installed into Chrome's data directory. On your next browser restart, suddenly you have AI features you didn't enable and a model taking up space you didn't allocate. Google's documentation mentions the feature, but most users don't read patch notes.

The trade-off calculus is interesting. Google gets better engagement with AI features if they're installed by default - most users won't seek them out manually. Users get AI capabilities they might actually use - the writing assistance is legitimately useful. But nobody asked if the trade-off was worth it.

For users on limited storage devices - Chromebooks with 64GB eMMC, older laptops, tablets - 4GB is a meaningful chunk of capacity. For users on metered connections, downloading a 4GB model without consent could cost real money. For enterprise environments with strict data governance, having an AI model appear on company laptops without IT approval is a compliance nightmare.

Google's response has been that users can disable the features and remove the model. Which is true, but misses the point. Opt-in versus opt-out matters. Downloading gigabytes of software without clear consent isn't respecting user agency, even if the software is technically useful.

This is the pattern with consumer AI deployment. Companies decide the features you need, install the models whether you want them or not, and frame it as innovation rather than forced adoption. Microsoft did similar things with Copilot integration in Windows. Apple is building on-device AI into iOS. Nobody's asking users if they actually want this.

The technology is clever. The deployment is questionable. On-device AI is coming whether users choose it or not - apparently, that decision has already been made for us. Your storage space, your bandwidth, your battery life - all now allocated to AI features you didn't request.

The model works. The question is whether respecting user choice also works, or if we've collectively decided that's not important anymore.