Mining Malware Specifications through Static Reachability Analysis

The number of malicious software (malware) is growing out of control. Syntactic signature based detection cannot cope with such growth and manual construction of malware signature databases needs to be replaced by computer learning based approaches. Currently, a single modern signature capturing the semantics of a malicious behavior can be used to replace an arbitrarily large number of old-fashioned syntactical signatures. However teaching computers to learn such behaviors is a challenge. Existing work relies on dynamic analysis to extract malicious behaviors, but such technique does not guarantee the coverage of all behaviors. To sidestep this limitation we show how to learn malware signatures using static reachability analysis. The idea is to model binary programs using pushdown systems (that can be used to model the stack operations occurring during the binary code execution), use reachability analysis to extract behaviors in the form of trees, and use subtrees that are common among the trees extracted from a training set of malware files as signatures. To detect malware we propose to use a tree automaton to compactly store malicious behavior trees and check if any of the subtrees extracted from the file under analysis is malicious. Experimental data shows that our approach can be used to learn signatures from a training set of malware files and use them to detect a test set of malware that is 5 times the size of the training set.