The increasing complexity of 5G networks has created new challenges for cybersecurity, especially with the rise of IoT devices that can be targeted by attackers to spread malware. This work proposes a novel approach to detecting 5G malware traffic using a polymorphic network packet transformer and neural network models. The system is able to transform network packets into various polymorphic forms, making it difficult for malware to evade detection. The paper presents three machine learning models, a typical 1D Convolutional Neural Network (1D-CNN), a LeNet-5 and a Vision Transformer, for detecting malware traffic, along with a preprocessing method that automatically learns features without prior knowledge of malware activity or feature extraction from network data. The proposed approach is more efficient, resilient, and adaptable to evolving threats and protocols. The enhanced toolkit, called Polymorphic Network Packet Transformer, can extract embeddings from raw traffic load using an autoencoder network, enabling more accurate representations that may be applied in a commercial Intrusion Detection System (IDS) application, in a protocol agnostic manner. The results show that the proposed system achieves higher accuracy rates in detecting 5G malware traffic and provides a new approach of defending against 5G malware attacks, paving the way for elaboration in 5G-powered connectivity of autonomous vehicles ecosystem, and the upcoming 6G networks.