Comma.ai uses uploaded video from anyone[1] with a Comma device to train their models, so it is most definitely not just SoCal. I thought they had a few datasets available because they run competitions on them but I didn't have much luck digging up supporting links

1: I believe it's possible to opt out, and believe it's also possible to change the upload endpoint to a storage bucket that you control, if you do want the segments just not for Comma to have them