Surprise, surprise:

Sandboxing protects against many threats, but does not suddenly make people ethical.

(source: Maintainers Matter)

as well as:

I imagine sandboxing will be used as an additional layer of safety in most distros (eventually) but mostly for the purposes of protecting from honest mistakes.

This would be in-line with what OpenBSD does with pledge, and it seems that this concept works well against honest mistakes, and does not rely on the user making policy decisions. (How well these policies work can be seen by looking at what people install on their mobile phones, and even more by sniffing the network traffic of these seemingly harmless apps.)

Whether an app is asking for a permission for the reasons the user thinks or whether it's also using it for other purposes, is in fact, undecidable — both scientifically and pragmatically speaking. (You can't give network access to a turing machine and still reason about that it actually only loads the weather forecast.)