<h2>It became a controlled technology when the risk became greater than the reward</h2>

> Why, in that situation, would the government NOT use it for mass
> surveillance?

Fifteen centuries is a LOT of time for people to find and execute ways of exploiting AI driven surveillance systems.  By then they will have been used by school shooters to ID and track targets, by terrorists to assassinate world leaders using smart weapons, they will have been used to automatically steal people's identities by reading information off of things in their hands, military secrets have been leaked because someone decided to proof-read a report while waiting at a bus stop. They will have been used to automatically find behavioral patterns that suggest people are cheating on their spouses so that you can blackmail people you've never even met, there are even websites where you can just type in someone's name, and the system will start sampling video of them to generate pornographic deep fakes or sex bots of them.  As the technology becomes more prolific, the number of people with the knowledge base to exploit it grows until the harm it does is clearly more detrimental than any gain.  

As the harmfulness of surveillance increases; so to will its utility decrease.  Once a deep fake can no longer be differentiated from real footage, shady defense lawyers just need to start submitting their own versions of events to cast doubt on any video or photos submitted by prosecution and vise versa. Heck, this is so far in the future, criminals probably walk around with portable holographic projectors that can create a bubble of false narrative without ever needed to hack a single system. Together, these will lead to video and photos becoming inadmissible as evidence. 

Then there is the issue of whether it makes cops more effective as a response tool.  Most "crimes" go unreported because no one wants to press charges.  If a person forgets their keys and has to break into their own home, there is no reason for the cops to show up.  If a husband hits his wife and the wife does not want to press charges, there's not much the cops can do.   If a person causes damage to property that is worth less than a lawsuit, then you are just wasting a cop's time filling out the paperwork.  And that is just the time you waste before you account for forged footage.  If you can only budget for so much law enforcement, then it makes more sense to dispatch them to places where you expect their efforts to result in a conviction which means going places where crimes have been reported by a person, not just a nosy AI. 

Between these factors, video surveillance becomes more of a hindrance than a help as a crime enforcement tool.

In short, public surveillance systems will become the 36th century equivalent to leaded gasoline, asbestos, or blood letting.  The tech might still be there in certain contexts, but the very idea that anyone ever thought it's widespread use was at one point a good idea seems borderline satirical.  

> Also how to stop private people and corporations from using it apart
> from outright banning it?

The best way to limit it without banning it is to making it something you need a license or permit for.  This way, the right to place a surveillance cameras becomes more akin to owning a concealed firearms license.  It's not that hard to aquire, but there is enough cost and red-tape associated with it that you start to only see cameras where they are actually needed.