Two years ago, GreenBorder, one of the early "sandbox" browsers, received mighty applause from Wall Street Journal tech guru Walt Mossberg. The sandbox browser -- basically, a browser running in a virtual container -- promised to keep nasty code from spilling into a computer's operating system and wreaking havoc.

Tests showed a year earlier that such sandbox technology didn't really work, at least not then. This year, to see if anything had changed, a new version of this technology was tested, in the form of Check Point's ZoneAlarm ForceField. Same result: it didn't do the job. "Within less than a minute, by clicking only my third test malicious Web site link, my test system was silently compromised without so much as a chirp out of ForceField," wrote security guru Roger Grimes.

The problem, then and now, is the sandbox wall remained permeable, so Trojans and other forms of malware can slip through the virtual sandbox into your desktop.

Grimes believes the technology approach of the sandbox browser, which he calls virtual red-green computing because the browser is the unsafe red zone and the operating system is the safe green zone, is fundamentally flawed -- and so he debunks the entire product category. Virtual red-green computing is the decades-old practice of creating separate safe and unsafe zones on a computer using virtualization techniques.

Sandbox browsers are largely aimed at consumers and small businesses. Products include ForceField (which claims to protect more than 60 million PCs), Sandboxie, Trusteer, and Safecentral.com. (Google bought GreenBorder a year ago, and it seems to have since disappeared as a product.)

Grimes' gripe: blurring red and green lines

The problem with sandbox browsers, Grimes says, is that eventually a browser and the operating system have to talk because they are so heavily integrated in the desktop environment -- that is, the red and green areas of a computer must co-mingle. It's nearly impossible to isolate one from the other, given browser plug-ins, JavaScript, and ActiveX controls that need to run code on the operating system.

In Grimes' ForceField test, for example, he visited known malicious Web sites with an unpatched Windows XP computer running ForceField and, again, with a fully patched computer sans ForceField. He did this to single out ForceField's capabilities. The fully patched computer foiled all attacks, whereas the non-patched computer with ForceField was breached on the third Web site Grimes visited.

The breach occurred when Grimes clicked on a malformed Flash file that caused a buffer overflow, sending an executable string of code into the computer's memory. "The malformed file used an API that was unexpected by ForceField to install itself as a service," he says. "Most of these limited emulation products cover dozens of the most popular APIs, but malware uses hundreds. Unless you cover every API call, [malware] is going to get around it."

In other words, "if the browser needs to launch an application, then all bets on the sandbox are off," says Doug Dineley, executive editor of the Test Center. "The application has to interact with the OS."