Detecting the Network Behavior of Malware

Roberto Perdisci - Dambala Labs / Georgia Tech
Digital Security

Date: June 1st 2010
Location: Eurecom - Eurecom

Most modern cyber attacks are carried out via malicious software (a.k.a. malware). It is therefore important to identify whether a machine has been compromised by malware, so that system and network administrators can take action to remediate the infection and prevent future attacks. Unfortunately, the battle against malicious software is becoming harder and harder. Today's malware developers commonly employ executable packing and other code obfuscation techniques to generate a large number of polymorphic malware variants. Existing anti-virus (AV) techniques are not able to effectively cope with obfuscated malware, thus leaving our computers vulnerable to malware infections. In this talk I will present the results of my research on detecting the network behavior of malware. The key observation is that most malware need to generate network activities in order to perpetrate their malicious intents. In addition, while code polymorphism allows for easily creating large numbers of variants of the same malware sample, when executed these variants will behave in a similar way because they share the same intended malicious goals and activities. I will show how we can automatically identify families of malware that share similar network behavior, and how we can model their malicious network activities to enable the detection of malware-compromised machines within a monitored network. I will also empirically demonstrate that network-based detection of malware behavior can complement traditional AV tools and other system-based malware clustering and detection approaches, thus representing a valuable part of a defense-in-depth strategy to protect computer networks from malware attacks.

Permalink: https://www.eurecom/seminar/17429