21 October 2025 in IEEE Access
As digital transformation accelerates, cyberspace has become increasingly active, resulting in a rise in cyberattacks. In particular, Advanced Persistent Threats (APTs) targeting high-value assets are difficult to defend against with conventional security systems due to their stealthy and persistent characteristics. This paper proposes a proactive defense framework for APT attacks that enables real-time responses by observing attacker behavior in near real-time and predicting subsequent attack steps. Unlike traditional methods that detect isolated attacks at individual security points, the proposed framework holistically observes attacker actions and constructs a provenance graph by linking correlated events. An attack scoring mechanism is applied to the graph, and once the score exceeds a predefined threshold, the activity is classified as an APT attack, prompting immediate response actions. Additionally, the framework learns attack technique patterns from over 1,300 past cyberattack campaigns to predict the next likely attacker behavior with a certain level of accuracy. This prediction allows for estimating the timing of attacks on victim systems and determining the optimal timing for defense measures. Experimental evaluation using six APT scenarios from the MITRE Evaluation demonstrated that the proposed system reduced the attacker’s dwell time on victim systems by 67% and effectively blocked APT progression. Furthermore, the framework outperformed conventional Endpoint Detection and Response (EDR) solutions.
30 September 2026 in The Journal of Korean Institute of Information Technology