The rise of containerized workloads in cloud-native environments has driven the need for more dynamic and scalable access control mechanisms. Traditional Role-Based Access Con- trol (RBAC) systems, while effective in static environments, face limitations when applied to highly dynamic cloud-native architec- tures such as Kubernetes. This paper explores the evolution from traditional RBAC to machine learning-driven adaptive security models. We propose an architecture that leverages anomaly detection and user behavior analytics to enhance security for con- tainerized workloads. Our approach enables real-time adaptation to evolving threats and user behaviors, addressing the challenges posed by dynamic cloud infrastructures. Comparative analysis demonstrates the superior adaptability and security performance of the proposed model over conventional RBAC systems. The results underscore the potential of integrating machine learning into access control, offering a robust solution for the security needs of modern cloud-native applications.