What Users Found
It started when someone looking through disk usage found several gigabytes of model files added to Chrome's config directory, with Gemini Nano already in place and callable by web-page APIs, while they had never clicked any consent button. Google's position is that this is a browser component update, no different from upgrading the rendering engine. Critics don't buy it: a rendering engine doesn't answer questions, and the arrival of a local AI model changes the boundary of the browser's capabilities—completing a change of this magnitude silently under the name of a "component update" doesn't hold up procedurally.
The Consent Problem of On-Device AI
This dispute previews a tension that will soon become common: vendors are all stuffing on-device models into operating systems and browsers, and logically "running locally is more private," but the deployment method is all act-first-report-later—the costs in disk, memory, and battery are passively borne by users, and turning it off means digging into experimental settings. The consensus in the comments is that the problem isn't the model running locally, but the old playbook of "on by default, switch buried deep" being applied to a new thing. And the browser happens to be the software users are most sensitive about—this pop-up wasn't worth skipping.
via: Hacker News