If only there was "article" you could read that might tell you. From TFA: The same researcher that took Google's money last year for exploiting Chrome, known publicly only as 'PinkiePie' was awarded $40,000 for exploiting Chrome/Chrome OS via a Linux kernel bug, config file error and a video parsing flaw. So, it sounds like Linux. Google fixed this by patching Chrome OS, not Chrome per se.

The browser is a rather complex beast and there is probably no way that the application itself can ensure system integrity... at least with any consistency.

Some of us are migrating our online activities to Qubes OS [qubes-os.org] which is a desktop distro (I know...) that allows you to create App VM domains for things like "personal", "work", "unsafe", etc. and also a "disposable" one that gets reset on exit. Each domain of apps is displayed in window borders that have an associated color.

Taking it further, some of the commonly-attacked system components like the network stack are virtualized as well.

Qubes employs VT-x and VT-d/IOMMU hardware to allow you to operate different types of peripherals (like bluetooth) without incurring all of the risk they normally carry. Even device drivers are paravirtualized! So the attack surface that can be used against the core system (or any other domains in the system) is kept to a bare minimum.

An added benefit of this approach is that user activities are tracked somewhat less than normal (especially if you use disposable VMs).

Chrome OS bug:
The CVE-2013-0913 hack was was a buffer overflow in the GPU for Chrome OS / Linux.
http://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2013-0913 [nist.gov]

Chrome browser bug:
Last year's PinkiePie hack chained multiple Chrome (browser) bugs together to be able to get to the GPU.
http://www.webpronews.com/google-chrome-cracked-by-six-bug-combo-2012-05 [webpronews.com]

They didn't release details yet, but odds are since it's the same person he probably used a similar method to hack the browser and get access to the GPU of the OS.

I don't know that it guarantees that it is a Linux problem. Did they modify the Linux source to do something specific for Chrome OS? Is the video parsing issue specific to Chrome OS? Did they do something non-standard or inherently unsafe with the config file?

Thats like arguing that your supermarket isnt to blame if they sell horsemeat as steak or something. You blame them (they are responsible for vetting the product they provide), and they will blame their vendor (who sold them a bad product).

As parent said, blame isnt this binary thing; multiple people can be at fault simultaneously, but as Chrome's vendor, the end user should look to Google for answers-- its not Linus' job to fix Chrome OS, its Google's.

You mean the same idea I've been asking for for about 15 years, otherwise known as bottling, process separation and lots of other fancy terms?

Your browser doesn't need access to the hard disk, except a single, solitary folder for downloads. That's it. It shouldn't even KNOW where that folder is, nor if it's in memory or a disk, or a network share. Hell, it shouldn't even be allowed to have the capability to look, let alone actually find out.

For uploads, the browser requests that YOU supply the information to the browser process bottle, and it takes it once supplied and does what's necessary. It has no need to have arbitrary access to every file visible on your system, only those it creates itself inside its bottle, or those you explicitly provide it with through some system mechanism. Similarly, it has no need to do anything more than put out a HTTP request and get a response.

Something else, somewhere, will handle, authorise and sanitise that request and response and do NOTHING else. Yes or NO. The program should have NO way to detect what that process is (so if the user wants to run in a zero-privilege environment, the browser just has to cope with that rather than say "I can't run without admin").

Now replace "browser" with "word processor", "spreadsheet", "hardware utility" or anything else that you use on your system.

The problem we have is that we've come from general purpose OS that were designed to let all processes have access to anything that wasn't explicitly locked away from them. The fix is to give processes the absolute bare minimum they require to do their work, make them ASK for everything, and refuse any request that you don't like. And make every process work (for the correct definition of work) even when tested inside a bottle that ALWAYS gets No to every request.

We've sort of tacked on such security features to today's OS (Unix-likes are certainly closer than, say, Windows), which historically always said "Yes", and now we have to start with one that says "No" all the time, for everything, and gives nothing to a process that isn't 100% necessary.

Replace all "file open dialog" actions with a system component that does NOTHING but let the user choose a file (Windows started out with the right idea here, but fails terribly in implementation). Hell, theming is then permanent and to the user's preference (and the program needs know NOTHING of the theme chosen or anything else) and nobody has to (or can) run around recreating an official file-open dialog. You can even "green-bar" official file-open dialogs (like we do with padlocks on SSL sites) so that they are distinguishable from rogue processes trying to create fake file-open dialogs (even though those would not be able to escape the bottle to read files anyway!). Make it so that NO other process can green-bar a file-open window except the file-open process.

Hell, why should a process even be able to know or change whether it's full-screen, windowed, the window size, etc.? Instantly you take ten options out of every game that has "recreated" those options and decisions for you and leave it to the user to decide. Game X will ALWAYS load fullscreen. Any process marked as a "Game" will only be fullscreen when I press this button. Or even "No process can EVER go fullscreen because I always like to see my Start Bar". And the process will have no way to know, and no way to override the decision of the user. All it knows is that it has a bitmap area it can draw to which is copied to the screen when it asks. It can't tell if that copy is a copy-and-scale into a window bitmap, or direct copy to video memory, or even just copy to a screenshot / VNC program.

Assuming a program wants to open a file, the program calls the function to open said dialog and is blocked until it returns. It can't do anything else but request it. The dialog is run in a process all of its own and has access to read file names in user-allowed folders, display things in a file-open dialog on-screen (again, subj