Abstract Vulnerability management remains a cornerstone of enterprise cybersecurity, but traditional approaches often fail to prioritize threats in alignment with real-world attacker behavior. This research proposes a novel automated system that uses deep learning to map publicly disclosed vulnerabilities (CVEs) to adversarial techniques defined in the MITRE ATT&CK framework. The model uses a combination of natural language embeddings and a multi-label deep neural network to predict potential attack techniques from CVE descriptions. Furthermore, it incorporates a risk prioritization mechanism that enhances decision-making beyond CVSS scores by considering behavioral threat context. Evaluation results demonstrate high prediction precision and acceptable recall, suggesting practical applicability for SOCs, threat hunters, and vulnerability analysts.
Abderehman et al. (Tue,) studied this question.