The technique abuses a legitimate way to control where a browser sends its requests, known as proxy auto-configuration or PAC, to take over a victim's browser and send traffic--say, requests to a bank--to an attacker-controlled server instead. While the attackers still have to find a way to execute code on a victim's system, once that is done, they can set a proxy for the browser, capture selected traffic, and re-route it invisibly.
"You can essentially have the end users interact with a shadow Internet, essentially an Internet experience that is designed by the bad guys," says Daniel Ingevaldson, chief technology officer for Easy Solutions, a fraud-prevention company that operates extensively in South America. "The victims surf normally, and it's completely undetectable to the end user, except for when they hit a site that is specified by the attacker."
In general, there is no indication in the browser that anything untoward is happening, Ingevaldson says. The proxy settings can be changed, for example, to only route requests intended for a single bank to an attacker's system. Those requests could also be modified in real time, he says.
"We haven't seen anything that sophisticated yet, but I can see a scenario when this attack can be used to implement a server-side or cloud-based man-in-the-middle attack," Ingevaldson says.
Most attacks have used the technique to create a phishing scheme, routing victims to a fake banking page that collects their credentials.
Security firms first reported the use of the PAC technique in 2010, but Brazilian cybercriminals have been using the technique for at least a few years before that, Ingevaldson says.
[DNSChanger shows that funneling infected network traffic to central servers can enable massive fraud, but the technique has significant weaknesses, as well. See Malicious Proxies May Become Standard Fare.]
In an analysis of one attack, cloud security firm Zscaler detailed a PAC file that would set a victim's browser to forward traffic for several Brazilian banking sites and American Express's site to an attacker-controlled server.
"What I have seen to date has been in Brazil," says Michael Sutton, vice president of research for Zscaler. "I expected to see it more widely used already."
The attack resembles DNSChanger, the widespread malware that changed a specific file on victim's systems to reroute all domain name lookups to an attacker-controlled server. With a PAC file, the attackers can be a lot more choosy, redirecting requests to a group of sites or to one site in particular, says Anup Ghosh, co-founder and CEO of Invincea, a firm which detect and blocks Web and e-mail threats.
Preventing PAC files from compromising browsers is not a simple task, as client-side security software will likely find it difficult to detect whether a give PAC file is a valid change or a malicious attack, he adds.
"The writers of Zeus and SpyEye are in the business of making money with the toolkits--so they are constantly changing tactics," Ghosh says. "Detecting that a change in the proxy is malicious is really hard from an anti-malware perspective, so that is a tough one for the [antivirus firms] of the world to address."
Yet the changes that malware makes to a system's registry keys to set the proxy settings could be detected, Zscaler's Sutton says. And if the compromise is being used to set up a man-in-the-middle attack, an observant security team at a bank should be able to detect the fraud.
"On the server side, I could tell that someone was connecting to me and I could tell that it was a proxy," Sutton says.
In addition, companies that use a proxy for security or DNS--and so use a proxy auto-configuration file to configure their employees' browsers--have a good chance of detecting changes that affect their infrastructure.
Have a comment on this story? Please click "Add Your Comment" below. If you'd like to contact Dark Reading's editors directly, send us a message.