"Should be closed-loop" is an interesting assumption that I'm not convinced aligns with the reality of the technology. To what extent has speech recognition been improved by being able to feed it through a constantly-updated set of open-loop ML infrastructure?
Probably a lot; open-ended voice recognition with no prior training is hard (though I'm not convinced it can't be made to work off-line now that the models are there). Still, a lot of devices don't need open-ended voice recognition (structured grammars greatly simplify the problem), and if you allow for users having to pre-train their devices for few minutes, off-line processing becomes easier.
My impression is that the main driver behind cloud-first voice processing is business, not technology.
In this particular case (speech recognition), one harm done to the users is tying the product to the Internet, and thus requiring what should be closed-loop tasks to go through vendor's servers.