The domain names used in espionage and cyber-crime operations hold clues to detecting the attacks before victims’ systems are compromised, researchers with Internet-security firm OpenDNS said on March 5.
The company, which processes tens of billions of domain-name requests every day, has developed an algorithm that uses a variety of natural-language processing techniques and other data to predict whether a domain is being used for malicious purposes. Dubbed NLPRank, the algorithm has successfully identified links used to infect victims in a number of malware campaigns, such as DarkHotel and Carbanak, according to two online posts describing the research.
When researchers study the command-and-control infrastructure of malware campaigns, they intuitively see patterns in the malicious domains, Andrew Hay, director of security research for OpenDNS, told eWEEK.
“If a domain was registered in Kuala Lampur and hosted in Russia, and has no connections within the network infrastructure to a company, that shows that there is something bad going on,” he said.
While OpenDNS can block domains known to be used in malware, the research effort aims to create a system that pre-emptively blocks links that have a high likelihood of being malicious. Already, OpenDNS is using the system to flag suspicious domain lookups, which are then routed to an analyst for further investigation. The system issues warnings for approximately 1,000 domain lookups every hour.
In many ways, the algorithm goes through the same logic as a security analyst, but on a far broader scale, taking into account a wide variety of factors. Jeremiah O’Conner, the OpenDNS researcher whose work led to the current system, thought up the technique while looking at the domains used in known malware campaigns.
“Typically, malicious domains fall into the pattern of using common ”abuse” words, which is why we decided to use natural-language text-processing techniques for this experiment,” O’Connor stated in a blog post. “One of the techniques we used to analyze these domains was to extract all the words found in the English dictionary and try to find any commonalities.”
The DarkHotel group, which infected targeted executives after they connected to hotel networks, used many domains that included certain words or word roots, including “auto,” “updat,” “serv,” and “free.” Domains related to news are also common in espionage operations conducted by the DarkHotel group, also known as APT–1. Domains such as “myyahoonews,” “newsesport,” and “newsonlinesite” were all used in that campaign.
OpenDNS researchers combined Whois records, geo-location data and information about the logical routing infrastructure, along with natural-language text processing, to find ways that the DarkHotel group’s domains were different.
The analysis has allowed OpenDNS to detect new domains used as part of attackers’ infrastructure, but until the company reduces the number of false alerts, also known as false positives, it will use analysts to manually review any suspicious domains.
“The research has not reached a place where we are automatically blocking whatever shakes out,” OpenDNS’ Hay said. “We are doing manual spot checks and adding to the block list as appropriate.”