A old browser attack that everyone thought was dead and buried is back, according to a security researcher at the Black Hat conference in Las Vegas.

Dan Kaminsky, director of penetration testing for IO Active, showed how problems in the way browsers handle domain name system requests could be exploited to give attackers access to any resources behind the corporate firewall.

He described a multi-step attack that could be used to scan corporate networks for data or vulnerabilities. But at the heart of the attack is a 1996 paper by Princeton researchers showing how a Java applet could be used to access systems on a victim's network. "It's one of the few things that's actually come back from the dead," Kaminsky said.

The fundamental problem, according to Kaminsky, is in the way that web browser software decides how to trust other computers. This decision is based on the computer's domain, and that DNS information can be misused, Kaminsky said.

"It's a binding problem," he said during an interview after his talk. "They assume a value is not changing, but the attacker can change it whenever he chooses."

For the past year, security researchers like Kaminsky have increasingly warned how flaws in the security model of applications could be misused by attackers.

In February, security researcher Robert Hansen showed how a DNS-based attack called "anti-DNS pinning" could be used to give an attacker access to any data indexed by Google Desktop.

Hansen said that while Kaminsky's talk may not have disclosed previously unknown vulnerabilities, "it's probably one of the coolest implementations" of this type of attack.

In his talk, Kaminsky described how a malicious website could interact with a browser and - following a complex chain of back-and-forth data requests - ultimately gain access to other resources on the victim's network. Attackers would be able to access any resource available to the victim running the browser, he said. "If you can reach it, so can the bad guy."

He plans to post further details of his attack on his Doxpara website later this week.

In Kaminsky's scenario, an attacker would use a proxy server that would send data to the browser, ultimately using Adobe's Flash player to trick the browser into trusting the outside website as if it was a local network resource, say a printer. "The proxy gets to update the browser to speak the necessary flash to service the bytes being sent to the attacker," he said.