<p>The standard methods for cloud resource distribution and threat classification operate upon fixed configurations, which are not able to adapt to varying application demands or new security risks. For this reason, the current methods suffer from issues such as ineffective resource allocation, unsatisfactory threat classification, and poor access control functionality with slow response time. Thus, the given paper introduces a hybrid framework comprising a deep convolutional network (DCN) equipped with a transformer module and a role-based access control (RBAC) mechanism to address such issues. The anomaly detection capability of the proposed framework derives from DCN for spatial feature extraction and the transformer module for temporal dependency analysis by managing resources at the same time. A streamlined pre-processing pipeline inside the CloudSim environment constitutes the proposed framework to achieve effective data ingestion quality and system performance for many workload types. Multiple benchmark datasets, such as CICIDS2017, Alibaba Cluster and Google Cluster, demonstrate the simulation capability of the framework. The results show that the proposed model achieves 97.6% accuracy and 95.8% F1 score for threat classification in the CICIDS2017 dataset. The Google Cluster shows 84% resource utilization with 68 ms latency discharge, and the Alibaba Cluster dataset achieves 82% utilization with 47 ms latency. The dynamic RBAC system maintains a success rate of 98.2% while achieving policy adaptation latency of 50 ms to tackle essential access control problems.</p>