Multi-Stage Attack Detection via Kill Chain State Machines.

Today, human security analysts collapse under the sheer volume of alerts they have to triage during investigations. The inability to cope with this load, coupled with a high false positive rate of alerts, creates alert fatigue. This results in failure to detect complex attacks, such as advanced persistent threats (APTs), because they manifest over long time frames and attackers tread carefully to evade detection mechanisms. In this paper, we contribute a new method to synthesize attack graphs from state machines. We use the network direction to derive potential attack stages from single and meta-alerts and model resulting attack scenarios in a kill chain state machine (KCSM). Our algorithm yields a graphical summary of the attack, APT scenario graphs, where nodes represent involved hosts and edges infection activity. We evaluate the feasibility of our approach in multiple experiments based on the CSE-CIC-IDS2018 data set. We obtain up to 446 458 singleton alerts that our algorithm condenses into 700 APT scenario graphs resulting in a reduction of up to three orders of magnitude. This reduction makes it feasible for human analysts to effectively triage potential incidents. An evaluation on the same data set, in which we embedded a synthetic yet realistic APT campaign, supports the applicability of our approach of detecting and contextualizing complex attacks. The APT scenario graphs constructed by our algorithm correctly link large parts of the APT campaign and present a coherent view to support the human analyst in further analyses.