A security system undergoing testing by a San-Francisco-based company aims to speed up the detection of websites and domains used for cybercrime.
The technology is being developed by OpenDNS, which specializes in performing DNS (Domain Name System) lookups. The DNS translates domain names such as idg.com into an IP address that can be called into a browser
OpenDNS offers a secure DNS service for ISPs and organizations that blocks requests from Web browsers to sites that may be associated with cybercrime or spoof a company such as PayPal.
The company, which was founded in 2005, has grown so much that its systems respond to some 71 billion DNS requests per day. That’s just 2 percent of global DNS traffic but is enough of a sample to pick up on many cybercrime campaigns.
The new system, called Natural Language Processing rank (NLPRank) looks at a range of metrics around a particular domain name or website to figure out if it’s suspicious.
It scores a domain name to figure out if it’s likely fraudulent by comparing it to a corpus of suspicious names or phrases. For example, g00gle.com—with zeros substituting for the letter “o”—would raise a red flag.
Many cybercriminal groups have surprisingly predictable patterns when registering domains names for their campaigns, a type of malicious vernacular that OpenDNS is indexing. Bogus domain names use company names, or phrases like “Java update,” “billinginfo” or “security-info” to try to appear legitimate.
But there’s a chance that NLPRank could trigger a false positive, flagging a variation of a domain that is legitimate, said Andrew Hay, director of security research at OpenDNS.
To prevent false positives, the system also checks to see if a particular domain is running on the same network, known as its ASN (autonomous system number), that the company or organization usually uses. NLPRank also looks at the HTML composition of a new domain. If it differs from that of the real organization, it can be a sign of fraud.
NLPRank is still being refined to make sure the false positive rate is as low as possible. But there have been encouraging signs that the system has already spotted malware campaigns seen by other security companies, Hay said.
Earlier this month, Kaspersky Lab released a report on a gang that stole upwards of US$1 billion from banks in 25 countries. The group infiltrated banks by gaining the login credentials to key systems through emails containing malicious code, which were opened by employees.
Hay said Kaspersky approached OpenDNS before the report was published to see if it had information on domains associated with the attacks. NLPRank was already blocking some of the suspicious domains, even though OpenDNS didn’t know more details about the attacks.
“We caught these things well back,” Hay said.
In some cases, NLPRank could allow a domain to be blocked even before one is actively used. After cybercriminals register a domain, they’ll often visit it once to make sure it’s accessible. It may then go dormant for a few days before it is incorporated in a campaign, Hay said.
If a fraudster is connected to an ISP that uses OpenDNS’s service, just a single DNS query for that new domain would allow OpenDNS to analyze and potentially block it before it is used for crime.
“As soon as we see that little bump on the wire, we can block it and monitor to see what’s going on,” Hay said. “It’s almost an early warning system for fraudulent activity.”