The rapid growth of software vulnerabilities has created a major challenge for organisations responsible for protecting critical infrastructure because conventional vulnerability management approaches frequently generate remediation queues that exceed available technical and human resources. This study developed and empirically evaluated an Artificial Intelligence (AI)-based predictive vulnerability management framework for prioritising vulnerabilities according to their probability of exploitation, asset criticality, exposure, and potential attack-path relevance. A controlled simulation-based experimental design was adopted using a synthetic dataset of 20,000 vulnerability records containing vulnerability severity, Exploit Prediction Scoring System (EPSS) score, Known Exploited Vulnerability (KEV) status, public exploit availability, asset criticality, internet exposure, attack-surface characteristics, asset age, patch complexity, dependency count, attacker interest, and mitigation strength. Four machine learning algorithms, namely Random Forest, Support Vector Machine, XGBoost, and LightGBM, were trained and evaluated using accuracy, precision, recall, F1-score, ROC-AUC, and false-positive rate. The results showed that LightGBM achieved the highest predictive accuracy (78.50%) and ROC-AUC (0.8090), while Random Forest produced the highest recall (94.35%). Comparative baseline analysis showed that the AI models provided stronger balanced predictive performance than CVSS-only, EPSS-threshold, and KEV-only approaches. The study further developed a context-aware prioritisation mechanism that combines predicted exploitation likelihood with asset criticality, internet exposure, and confirmed exploitation evidence. The findings demonstrate that AI can support a transition from severity-based vulnerability management to predictive and context-aware remediation. The proposed framework provides a practical basis for identifying vulnerabilities most likely to be exploited, prioritising critical systems, and informing attack-path-oriented defensive decisions within critical infrastructure environments.