In recent years, the significant rise of cyber threats has necessitated improved anomaly detection systems to secure digital assets and infrastructure. In this project, we are devising a system which make use of a few artificial intelligence (AI) and machine learning (ML) techniques for anomaly detection which recognize a unusual behavior of an entity and take action to stop the attack when it is happening. Utilizing ML algorithms, including clustering, classification, and even deep learning models, the system can accurately analyze network traffic, user behavior, and system logs to identify anomalies across massive datasets. It explores various methods of SMOTE and hybrid learning, determine key challenges faced with using machine learning for cyber security, including but not limited to data imbalance, false-positives with a novel focus on the dynamic nature of cyber security threats and the robustness of the solution. The proposed solution is a way to improve what such systems are already capable of identifying by combining advanced AI algorithms such as transformer-based models but paired with extensive preprocessing and feature engineering. This study aims to demonstrate the use-cases of AI-driven anomaly detection to strengthen cybersecurity resilience while also providing an outlook into how these systems can operate within real-world applications along with challenges to ensure scalability and real-world efficiency and reliability.