That said, personally, I think it's unlikely.
First, the vast majority of AI research is happening out in the open. It's often funded by militaries (e.g. DARPA) but it's not owned by them directly, rather the owners are universities and similar.
Second, private corps promote their AI advances very aggressively and this marketing is a big part of the advantage they try to establish over their competitors, so it doesn't seem obvious why they would keep a new advance "under wraps". For example, what would Google do with a secret machine vision system capable of 20% better performance than that of its rivals? I can't tell.
Of course we can assume that Google or someome similar have a much more advanced system than anyone else, like a real AGI. We can imagine such a system sitting at the heart of Google, dirctly advice Larry and Sergey how to run the company. But then, if we can imagine that, we can imagine anything.
And I do mean anything:
https://gatherer.wizards.com/Pages/Card/Details.aspx?multive...
As in security cameras, finances, cell tower data, etc. Facial and gait recognition, location tracking, etc. far beyond anything commercially available. But I'd doubt they have an AGI behind closed doors :P
Perhaps it's not the best idea to use these more advanced Level 4 systems in the densest city yet, but most of driving does usually happen on highways and regional roads so it's not a major loss.
I'm sure someone will eventually figure out how to detect "eyyy im walkin eer" and honk back automatically, and then it should be usable in NY as well.
> versions of AI that are hidden from the public both private and military
I do think you overestimate how much the military has its shit together. These systems need gigantic amounts of driving data to train and nobody else has that aside from car manufacturers with their fleets. And out of those Tesla is the one that's been doing it the longest.