With the advent of 6G technology, communication networks are predicted to enable ultra-high data transmission rates, extremely low latency and highly reliable connectivity. A key enabler of these innovations is network slicing, which allows numerous virtualized slices to coexist on the same physical infrastructure, each customized to specific service requirements. However, this flexibility also presents new security concerns. In this paper, we propose a slice‑specific attack classification system that employs artificial intelligence to detect important threats, including spoofing, distributed denial of service (DDoS), and eavesdropping. Using a balanced dataset of slice kinds (Ultra‑High‑Speed, Autonomous, IoT) and attack categories, we train and test machine learning and deep learning models to analyze traffic patterns and classify attacks. Our results show both the potential and limitations of classical classifiers such as Decision Trees and Random Forests, which produce high training accuracy but suffer from overfitting and poor test performance. These findings underscore the future demand for lightweight, adaptive, and federated learning techniques to offer scalable and real‑time security. Overall, our research contributes to the design of secure 6G slice architectures by providing a baseline foundation for attack detection and a path for future research in adaptive AI‑based defense systems.