The Fifth Generation (5G) specifications have set a precedent for the evolution of next-generation mobile networks. Standardized interfaces and Network Function Virtualization (NFV) technology enable network operators to break free from vendor lock-in, while delivering more customized and agile services to their customers. However, the heterogeneous and multi-vendor composition of the Next-Generation Network (NGN), as envisioned in 5G specifications, also expands the existing attack surface and complicates trust relationships. Consequently, the traditional perimeter-based security model has become inadequate for effectively ensuring trust in such a complex network environment. On the other hand, Zero Trust has emerged as a promising security model well-suited for protecting complex and large-scale networks. Unfortunately, the current access control mechanism in the 5G core network lacks key features, rendering it incompatible with Zero Trust principles. To bridge this gap, we introduce the Continual Access Monitoring (CAM) framework that enables operators to seamlessly incorporate key security metrics into the existing access control mechanism. Furthermore, CAM introduces continual access policy evaluation, a critical requirement of the Zero Trust paradigm. The CAM framework illustrates a practical strategy for integrating Zero Trust principles into the 5G service-based architecture and scales efficiently in large 5G deployments, supporting access policy monitoring for up to 6,000 network functions at an operational cost of USD 0.2 per hour on AWS.

The use of transformers for vision tasks has challenged the traditional dominant role of convolutional neural networks (CNN) in computer vision (CV). For image classification tasks, Vision Transformer (ViT) effectively establishes spatial relationships between patches within images, directing attention to important areas for accurate predictions. However, similar to CNNs, ViTs are vulnerable to adversarial attacks, which mislead the image classifier into making incorrect decisions on images with carefully designed perturbations. Moreover, adversarial patch attacks, which introduce arbitrary perturbations within a small area (usually less than 3% of pixels), pose a more serious threat to ViTs. Even worse, traditional detection methods, originally designed for CNN models, are impractical or suffer significant performance degradation when applied to ViTs, and they generally overlook patch attacks.In this paper, we propose ViTGuard as a general detection method for defending ViT models against adversarial attacks, including typical attacks where perturbations spread over the entire input (Lpnorm attacks) and patch attacks. ViTGuard uses a Masked Autoencoder (MAE) model to recover randomly masked patches from the unmasked regions, providing a flexible image reconstruction strategy. Then, threshold-based detectors leverage distinctive ViT features, including attention maps and classification (CLS) token representations, to distinguish between normal and adversarial samples. The MAE model does not involve any adversarial samples during training, ensuring the effectiveness of our detectors against unseen attacks. ViTGuard is compared with seven existing detection methods under nine attacks across three datasets with different sizes. The evaluation results show the superiority of ViTGuard over existing detectors. Finally, considering the potential detection evasion, we further demonstrate ViTGuard’s robustness against adaptive attacks for evasion.