Biblio

Data on software vulnerabilities, products and exploits is typically collected from multiple non-structured sources. Valuable information, e.g., on which products are affected by which exploits, is conveyed by matching data from those sources, i.e., through their relations. In this paper, we leverage this simple albeit unexplored observation to introduce a statistical relational learning (SRL) approach for the analysis of vulnerabilities, products and exploits. In particular, we focus on the problem of determining the existence of an exploit for a given product, given information about the relations between products and vulnerabilities, and vulnerabilities and exploits, focusing on Industrial Control Systems (ICS), the National Vulnerability Database and ExploitDB. Using RDN-Boost, we were able to reach an AUC ROC of 0.83 and an AUC PR of 0.69 for the problem at hand. To reach that performance, we indicate that it is instrumental to include textual features, e.g., extracted from the description of vulnerabilities, as well as structured information, e.g., about product categories. In addition, using interpretable relational regression trees we report simple rules that shed insight on factors impacting the weaponization of ICS products.

When created, the Java platform was among the first runtimes designed with security in mind. Yet, numerous Java versions were shown to contain far-reaching vulnerabilities, permitting denial-of-service attacks or even worse allowing intruders to bypass the runtime's sandbox mechanisms, opening the host system up to many kinds of further attacks. This paper presents a systematic in-depth study of 87 publicly available Java exploits found in the wild. By collecting, minimizing and categorizing those exploits, we identify their commonalities and root causes, with the goal of determining the weak spots in the Java security architecture and possible countermeasures. Our findings reveal that the exploits heavily rely on a set of nine weaknesses, including unauthorized use of restricted classes and confused deputies in combination with caller-sensitive methods. We further show that all attack vectors implemented by the exploits belong to one of three categories: single-step attacks, restricted-class attacks, and information hiding attacks. The analysis allows us to propose ideas for improving the security architecture to spawn further research in this area.

A frequent claim that has not been validated is that signature based network intrusion detection systems (SNIDS) cannot detect zero-day attacks. This paper studies this property by testing 356 severe attacks on the SNIDS Snort, configured with an old official rule set. Of these attacks, 183 attacks are zero-days' to the rule set and 173 attacks are theoretically known to it. The results from the study show that Snort clearly is able to detect zero-days' (a mean of 17% detection). The detection rate is however on overall greater for theoretically known attacks (a mean of 54% detection). The paper then investigates how the zero-days' are detected, how prone the corresponding signatures are to false alarms, and how easily they can be evaded. Analyses of these aspects suggest that a conservative estimate on zero-day detection by Snort is 8.2%.