Severity-based triage of cybersecurity incidents using kill chain attack graphs

Abstract

Security teams process a vast number of security events. Their security analysts spend considerable time triaging cybersecurity alerts. Many alerts reveal incidents that must be handled first and escalated to the more experienced staff to allow appropriate responses according to their severity. The current state requires an automated approach, considering contextual relationships among security events, especially detected attack tactics and techniques. In this paper, we propose a new graph-based approach for incident triage. First, it generates a kill chain attack graph from host and network data. Second, it creates sequences of detected alerts that could represent ongoing multi-step cyber attacks and matches them with the attack graph. Last, it assigns severity levels to the created sequences of alerts according to the most advanced kill chain phases that were used and the criticality of assets. We implemented the approach using the MulVAL attack graph generator and generation rules for MITRE ATT&CK techniques. The evaluation was accomplished in a testbed where multi-step attack scenarios were executed. Classification of sequences of alerts based on computed match scores obtained 0.95 area under the receiver operating characteristic curve in a feasible time. Moreover, a threshold exists for classifying 80% of positive sequences correctly and only a small percentage of negative sequences wrongly. Therefore, the approach selects malicious sequences of alerts and significantly improves incident triage.