Zero trust cloud architectures enhanced by predictive AI- based threat modelling

Cloud computing environments have increased the enterprise attack space making the traditional perimeter-based approach to security obsolete. Zero Trust Architecture (ZTA) requires the continuous validation of identities, devices, and contextual attributes but most of their applications are still reactive, rebalancing trust after anomalies in its observance have been detected. At the same time, predictive Artificial Intelligence (AI) systems produce probabilistic threat measurements, but are usually limited to monitoring capabilities as opposed to being implemented as a component of real-time access control decisions. The paper will present a Risk-Integrated Zero Trust Architecture (RI-ZTA) wherein it is formally assumed that externally generated risk scores based on predictive AI are introduced in the dynamic computation of trust. Trust has been modified to incorporate identity assurance, compliance of device, contextual integrity, and predictive probability of threat in the Policy Decision Point. Analytical analysis in controlled adversarial conditions shows quantifiable scores in comparison to conventional ZTA and perimeter-based frameworks. In particular, RI-ZTA shortens the Time to Trust Recalibration of 60 (standard ZTA) and 120 seconds (perimeter) to 25 seconds, shortens the attack propagation time by 140 seconds to 55 seconds, and decreases the policy adaptation latency by 45 seconds to 18 seconds. These findings demonstrate that predictive risk intelligence that is directly incorporated into trust computation allows threshold crossing to take place sooner, and reduced dwell time of attackers and faster enforcement judgments. The suggested system can further improve proactive containment within the cloud settings without the need to retrain the current AI detection systems.