This is the case on _every_ OS that lets users install applications.
So, how can we make sure users doing tasks aren't afflicted by applications that want to sneak in? You're saying to "structure a security model differently" leads me to think maybe users shouldn't have admin access on the machine? But then you get all the critiques people have of iOS.
I'm honestly asking, what's the alternative here?
Locking down the entire UX doesn’t work because day-to-day operations become onerous while the constant spew of “Are you sure?” dialogs teaches users to click OK without thinking.
Instead, put the sensitive stuff into its own “secure desktop”, and police all movements between that and the “standard desktop”. Even stick in an “unsafe desktop” for doing risky stuff, and encourage users to install new/untrusted software there; e.g. let them install and run anything without nagging, but keep apps completely sandboxed. Boundaries should be clearly delimited, e.g. give the high-security desktop its own window chrome and fixed background image that can’t be mimicked on the others.
It’s deliberately coarse-grained, unlike the current mess, and that’s the point. That makes it easy for users to mentally model and match scopes to tasks, and limits nagging to where it really matters: when crossing scope boundaries with data in tow. Simply switching scopes and working within a scope should not nag, however.
Look, I’m not saying it’s guaranteed to work. But at least I’m thinking about it from the point of view of end-users, not vendor developers and lawyers. It’d need prototyped and tested in both lab and real world, and sure as guns users will discover exciting new ways to screw things up; which means learning and adapting to account for that.
But that’s UX. Either do it right, or go get a job flipping burgers where the worst you’ll do is hospitalize a few dozen locals with E. Coli. TWO BILLION now trust large chunks of their lives to their “personal” machines. That is a lot of people to screw up.
Like Apple, you are thinking App-centrically. Stop. Apps are not the center of the users’ universe. The users’ own tasks are.
Therefore you need to think task-centrically—i.e. What does the user need/want to do?—and structure your security model to enable and support that.