Ransomware dynamics: Mitigating personal data exfiltration through the SCIRAS lens

Abstract

Ransomware’s capability to exfiltrate personal data is one of the most significant threats to privacy today. Its growing complexity and resistance to static analysis have driven research efforts to implement security controls on endpoints using dynamic analysis. However, the critical security threshold that these endpoint controls must overcome to effectively mitigate personal data exfiltration and stop ransomware propagation once an infection has begun in communication networks remains unclear. This paper addresses this issue by analyzing the Susceptible–Carriers–Infected–Recovered–Attacked–Susceptible (SCIRAS) epidemiological model in the context of a critical ransomware attack, with limited network and administrative security, that defines the critical scenario to be overcome. Unlike previous studies, this research first estimates a critical execution rate by studying the behavior of LockBit, Ryuk, and TeslaCrypt ransomware families and simulating CL0P MOVEit and Conti attacks in a controlled environment. To reflect more realistic conditions, we introduce a critical dynamic infection rate based on the critical execution rate, several attack vectors of modern ransomware, and the effect of limited network security. Using this baseline, a proposed triple extortion SCIRAS model is simulated and analyzed under its estimated parameters’ critical values to solve for each ransomware family the optimization problem of finding the critical security threshold required for endpoint controls to reach the Kermack and McKendrick’s non-epidemic status with the minimum feasible basic reproduction number. Our results demonstrate that a critical security threshold of at least 0.961 might contain modern ransomware exceeding the thresholds reported in previous simulations of SCIRAS and other models. Furthermore, we introduce a novel deep-learning-based framework called RansomSentinel, validated on the RanSAP120GB, RanSAP250GB, and RanSMAP datasets, which outperforms traditional machine learning classifiers and surpasses the estimated critical security threshold of each analyzed ransomware family.