Unfortunately, the nature of the world is such that if you allow users to control sandboxing, the next wave of attacks will come from applications that kindly ask users to disable the sandbox because of $reasons.
You can make the best locks in the world and it's all for nothing if every time an attacker knocks on the door the user opens it up and lets them in, but the solution can't be to weld shut every door.
Security commonly fails at UX. We could do better.
But at some point, if you ask the user "should this app access your private information" and the user says yes, that's what needs to happen, and the user needs to learn when to say no.
Linux evades a huge amount of virus and phishing just because you have to `chmod +x` stuff before running, for example. A "Software wants to do nasty stuff. Allow | Disallow" prompt just does not make it.
Also, users will want to piece the sandbox some times, do not make it an all or nothing situation. Make giving common permissions easy, rare permission hard, and give fine grained control over them.
True - but isn't the latter exactly the problem we're facing? That we increasingly don't get to decide whether we want that sandbox or not?