Incidentally this is exactly what Android does. They provide the isolation by creating a new user account for each app and by default apps get no permissions. The permissions you accept at install time then allow extra access. (Behind the scenes there are also mechanisms to provide security permissions across processes so that a second process doing work on behalf of your process is limited to your processes permissions.) Multiple problems have happened.
The initial default permissions in Android 1.0 were a little too permissive and for example allowed reading from the sdcard. While that seemed harmless at the time, it was in retrospect a bad choice.
There is no hard validity checker. For example if an application ignores the API call to find the location of the sdcard and instead just uses /sdcard then no one would know. This means that it is very easy for apps to not comply with best practises to be available, the developers don't even know it, and the platform has problems if they want to fix the issues due to the large installed base. (If Angry Birds stopped working because they hard coded things instead of using APIs would Android or Rovio get blamed?)
Android also made a decision that you either accept an app and its permissions, or not install it. There are good arguments both ways of only allowing them as a whole or being able to disable some permissions. But the net effect is that many apps have more permissions than they do use, and have to declare them if there is any chance they could be used. Users can choose to not install apps with too many permissions, but you can't leave a review explaining that. I see a lot more pushback over permissions in app store reviews these days, but the many years of prior practise mean far too many apps have too many permissions. (BlackBerry had this scheme where for each permission you could set it to grant, deny or ask on each use. For the tech crowd it is great, but for someone who just wants to play a game it is annoying and probably not helpful.)
The permissions themselves are not sufficiently granular - for example "Internet" is on or off. An email app contacting my email server is great, but contacting a command and control server in another country is a huge problem. Of course some sort of per connection prompting could happen, but isn't really practical for the vast majority of users.
Commercial issues have a been a problem too. For example the Google Play Store is not available in China (and many other countries). Consequently users have to use a different store, and some have different incentives as app stores are not lucrative.
Finally we have the "fragmentation" issue. Google did not implement Android in such a way that the majority of core could be upgraded by Google. Consequently you end up with full stack releases which depend on manufacturers and carriers deploying after they have already sold a device - something they have no incentive to do. This would need to be solved by aligning incentives, and is hard.
TLDR: the hard stuff is hard. Application isolation is already done by all the mobile operating systems. It is the nitty gritty details beyond that that need solving.
I don't like the user-account-per-app paradigm. It seems like an ugly abuse of multi-user access control to compensate for the missing app isolation piece.
I agree that it's a hard problem.
Preventing an app from opening other apps or arbitrary files if the user desires is awful for the users. Siloing data so that a user cannot use an arbitrary application to read a PDF is awful for the users. One-click installs and refusing to run unsigned apps is awful for the users when the only CA is Apple or Microsoft.
All of the current systems are poorly implemented. They could either be a step forward or two steps back. I'm not sure of the direction they are taking yet.
Remember software engineers have been stamping out bugs in the x86/64 arch for decades and they still find new bugs all the time. ARM arch is no different, they'll be stamping out bugs for decades too. By adding a hypervisor and virtual machines you're layering a whole new set of bugs on top like a bug sandwich. Complexity skyrockets, security is the first casualty.
As for not having root on a proprietary device it's a double edged sword. On one hand you're firmly in the land of feudal security leaving everything up to the developers, on the other hand if you root the device you're now opening up priv escalation and NFC exploits galore.
So we end up with shitty phones and tablets that try to make consuming old media easier, instead of users making their own media and applications.
There is also the Sandboxie approach to installing apps.
Once installed the OS monitors the app (and any spawned processes) for any and all changes it makes to the OS and once closed the settings are reverted (except local settings). When uninstalled any files/settings/etc it changed are also reverted by the OS instead of the uninstaller.
(1) Operating systems have an outmoded security model. Most focus on multi-user security, which still has its uses, but they fail to focus on application isolation. Applications should be installable by anyone and isolated completely from other apps unless specifically granted permission by a user/administrator. The entire malware problem can be laid at the feet of this.
(2) The (related) poor state of installability. Mac has this problem the least with drag-and-drop .app installation, though sometimes even that can be confusing (and .dmg packages are weird beasts... why?). Linux has .rpm or .deb, which applies a massive and complex band-aid to the otherwise awful state of installability on that platform. Windows is absolutely horrible... it's like Linux where you have "installers" that have to do package management instead of a formal package system.
The open source world -- and even commercial vendors that want to keep the PC alive -- have to either address this problem or accept the dominion of the locked-down vendor-controlled consumer compute device. This will require abandoning the old fashioned Unix design philosophy (and the similar way Windows works) and thinking seriously about the problems of installability and isolation. It would be worth taking cues from iOS and Android here, though there's also a lot of room for new ideas.
Oh, and I forgot to mention. If we don't address these problems, all app vendors will pay a ~30% per-sale tax to Apple, Google, and Microsoft in exchange for the valuable service of a platform that provides installability and application isolation. And you know what? The market will pay it, because for most users those things are that valuable.
Point, click, install, with no fear of damaging my system. If I don't like it I click and uninstall. Anything else is completely broken.
It's interesting to note that the prevalence of virtualization is also a sign of the failure of operating systems. OSes in a box (whether via complex container overlays like OpenVZ/Virtuozzo or hypervisors) are an ugly hack to fix the fact that the OS security model is broken even for multi-user operation. The fact that everything requires root to install is the deepest issue, along with the lack of permission structures for things like network interfaces. It should be possible to run an OS and sell accounts to the general public, not VMs, and people should be able to run whatever they want from their local account and this should be safe. The fact that this isn't viable is because OSes are broken, thus we have the huge overhead of virtualization as an ugly band-aid to fix it.