Supervised detection of infected machines using anti-virus induced labels

Traditional antivirus software relies on signatures to uniquely identify malicious files. Malware writers, on the other hand, have responded by developing obfuscation techniques with the goal of evading content-based detection. A consequence of this arms race is that numerous new malware instances are generated every day, thus limiting the effectiveness of static detection approaches. For effective and timely malware detection, signature-based mechanisms must be augmented with detection approaches that are harder to evade. We introduce a novel detector that uses the information gathered by IBM’s QRadar SIEM (Security Information and Event Management) system and leverages anti-virus reports for automatically generating a labelled training set for identifying malware. Using this training set, our detector is able to automatically detect complex and dynamic patterns of suspicious machine behavior and issue high-quality security alerts. We believe that our approach can be used for providing a detection scheme that complements signature-based detection and is harder to circumvent.