Sentine1X: Hybrid Human–AI Collaboration for Real-Time Web Threat Mitigation

Abstract

Modern web applications face increasingly sophisticated
threats that exploit application logic, API misconfigurations,
and behavioral patterns often missed by signature-based
defenses. While machine learning offers scalable detection, fully
autonomous systems lack transparency and may produce excessive
false positives or overblocking. In this paper, we introduce
SentinelX, a hybrid human–AI framework for real-time web
threat mitigation that integrates zero-trust policies, predictive
uncertainty, and explainable decision-making. SentinelX fuses
supervised classification and unsupervised anomaly detection
with dynamic trust scoring and SOAR-based response actions.
It incorporates SHAP/LIME explanations, human-in-the-loop
escalation, and an active learning loop to adapt to novel threats.
Evaluation on web traffic datasets and simulated attack scenarios
shows that SentinelX significantly improves detection precision,
reduces mean time to detect and respond, and minimizes analyst
workload compared to state-of-the-art baselines. The system
provides a practical, trustworthy blueprint for deploying safe
automation in modern security operations.