Black Hat: DLP Hack

Careful -- your data leak prevention tools may be, well, leaky.

A pair of researchers has discovered multiple types of flaws in various vendors' DLP products that would let an attacker evade them, alter their records of stolen data, and even use them to bot-infect client machines.

At the heart of the problem is the way some DLP products are being designed, the researchers say. Their underlying approach, a sort of honor system, is problematic: "It's like handing a bracelet to all the suspects and saying, 'don't do anything wrong, or we'll catch you,'" says Eric Mondi, a security researcher with Matasano Security who led the firm's research on DLP products, which will be presented at Black Hat USA next month in Las Vegas.

"By the time you've detected your most sensitive information was leaked, the ultimate value of the DLP product is [gone]" and the attacker has copied a dossier of data on your firm's Social Security numbers, says Thomas Ptacek, co-founder and researcher with Matasano Security. "Forget about calling the security team in: It's over. You need to call PR and try to mitigate" the publicity fallout, he says.

Matasano won't name names, but several of the DLP vendors the firm alerted about the bugs -- which include buffer overflows and SQL injection -- are already working on fixes. Even so, enterprises need to know that these tools can backfire if they're not secured or audited, the researchers say.

"DLP is a top line-item for IT," Ptacek says. "A vulnerability in the piece of software that controls hundreds or thousands of machines is a catastrophe... if an attacker can find that vulnerability and take control of it. If it’s not extremely well-audited, there [will be] latent botnet infections on your network."

Some DLP products are especially leaky. The communication between agent and server was weak in many DLP products the researchers tested. "It was mainly ad hoc and weak encryption," notes Ptacek. "This data is the crown jewels of the enterprise, and the communication between agent and server has to be [better] protected."

And with Windows machines, DLP products must embed code into the kernel, which of course opens another can of worms. "The [more] stuff that's loaded into the kernel, the harder it [the DLP product] is to evade," Ptacek says. "But it also exposes more vulnerabilities to the kernel itself."

Monti says he and Ptacek will demonstrate at Black Hat a fictional DLP product that basically combines typical DLP features (in addition to the common bugs) to illustrate the risks of these tools. The goal of the researchers is to make organizations aware of these vulnerabilities in DLP products, and to know how to spot them.

"One of the things we advocate is that they do their homework on all regulatory compliance criteria," Monti says. "If they are using DLP products that don't comply to that level... they are actually failing compliance, because they are using this security product," Monti notes.

Cory Scott, vice president of global research, guidance, and consulting at ABN-AMRO, says security tools shouldn't introduce any risk to the enterprise. "The technology is only as good as the implementation," says Scott, who was speaking independently, not on behalf of ABN-AMRO. "In the cases of the vendors that Tom and company looked at, it appears as if the development and design practices were lacking. Think of the Hippocratic oath: You don't want the cure to be worse than the disease."

Still, although these products won't stop a determined attacker, Scott says, if you properly vet and audit them, you're practicing due diligence in protecting your data as well as that of your customers.

— Kelly Jackson Higgins, Senior Editor, Dark Reading