Privacy
The scan runs in your browser. No frames are uploaded, there is no account, and the only network request is the object model, once.
Every frame from your camera is read, measured and discarded in the same moment. Frames are never uploaded, never written to storage, and never sent anywhere. No server sees your room, because no server is involved in the scan at all — the analysis is JavaScript running in your own browser tab.
Starting a scan downloads an object model, about 18 MB, from Google's public model host at storage.googleapis.com. That request is what lets the scan name the object a flagged point sits on. Like any web request it reveals your IP address and the fact that a file was fetched, to Google, and is subject to their handling of it. It happens once and the browser caches the result. No image, no frame and no scan result is included: the traffic goes one way, and what comes back is the model.
There is no account, so there is nothing to sign up for and nothing to delete. No analytics run on the scan page. No cookies are set for tracking — the only stored value is your language preference. No scan result is recorded, which also means nothing can be handed over on request, because nothing was ever kept.
Your browser asks before the camera opens, and it asks on your terms rather than ours. The stream is released the moment you end the scan, close the tab, or switch away from it. You can withdraw the permission at any time in your browser's site settings, and the scan simply stops working — it has no other way in.
If what the scan does with data changes, this page changes with it, and so does the screen you see before a scan starts. The download notice on that screen exists for exactly that reason: something that happens automatically should be stated before it happens, not buried here.