In the increasing complexity of cyber-physical backgrounds, ultra-reliable, smart, and self-sufficient security models are needed for this modern sixth-generation (6G) networks. Wide range of devices, low-delay services, and dynamic risks in 6G are not effectively managed by conventional centralized security systems. For seamless security managements in 6G backgrounds, the Federated Learning (FL) and Digital Twin (DT) Enabled Predictive Security Architecture (PSA) with Standardized Policy Enforcement was suggested in this study. A decentralized method named FL, is utilized by the suggested method. This application maintains data privacy and transmitting threat data over various network areas. Thus, data privacy is maintained, and communication cost is reduced. The real-time virtual copies of the network components are created by the implementation of DT. Through continual monitoring, anomaly detection, and applying proactive mitigation strategies, this DT offers predictive security. Collaborating various structures, quick adaptation, and satisfying the demands of several stakeholders are all ensured by a unified policy enforcement layer. The predictions became more accurate, resilient, and adaptable, and it was attained by integrating FL with DT. Through this integration, adaptable threat detection, and automated defensive orchestration is facilitated. In 6G, with the implementation of zero-touch network and service management (ZSM), self-healing, self-optimizing, and self-configuring capabilities are facilitated by the suggested system. Then, simulation was conducted to determine the efficiency of the suggested method. From the outcomes, it is clear that the suggested system attains low false positives (FP), fast response rates when compared to conventional methods. A scalable, smart, and policy-compliant predictive security model was presented in this study. More accurate, autonomous, and effective 6G backgrounds are facilitated by this policy-compliant predictive security model. FL achieved 94–96% detection accuracy, 2.6–3.0% false positives, 92–101 ms latency, 3.1–4.2 MB overhead, 90–94 compliance, while DT recorded 94–95%, 2.9–3.3%, 98–107 ms, 3.4–4.5 MB, 93–95, respectively. © 2017 IEEE.