AI-Driven Dynamic Risk Management in Cybersecurity

Abstract

Because of the growing technological sophistication of companies, the threat of attacks grows, and static risk assessments can no longer meet the requirements. AI-powered systems can provide real-time detection capabilities, but the "black box" nature of the technology makes it difficult to ensure their adoption and to justify their financial cost. This brings the need for transparency and explanations of how AI works in cybersecurity, so that organizations can better comprehend, validate, and rely on AI's intelligent decision-making. The survey assesses recent progress in explaining AI (XAI) in dynamic cybersecurity risk management in critical infrastructure systems such as 5G and banking. We examine techniques from LLM-based vulnerability prioritization to hybrid cascading risk models. This paper outlines how recent frameworks have leveraged multiple datasets, such as EPSS and CISA KEV, to enhance remediation efficiency by up to 95%. Unlike existing surveys, this paper provides a unified review of explainable AI, dynamic cybersecurity risk management, governance, and cyber insurance. It also highlights existing research on these fields, explores practical implementation issues like computational needs, data protection, and model generalization, and proposes future research avenues to create not only transparent but also trusted frameworks for mitigating cyberattacks using AI techniques. The analysis further demonstrates that XAI tools such as SHAP and LIME are essential for the financial health of cybersecurity. Lastly, attention must be given to changing the auditable logic to make technical evidence measurable and insurable.