The escalating complexity of cybersecurity threats necessitates advanced technological solutions to protect digital infrastructures. This study explores the application of Autoencoder neural networks, a deep learning model, for anomaly detection in network traffic, aiming to enhance real-time identification of cyberattacks. Using the CICIDS2017 dataset, which encompasses diverse attack types such as Distributed Denial of Service (DDoS) and infiltration, the Autoencoder was trained to detect deviations from normal traffic patterns based on reconstruction errors. The model was optimized through preprocessing, feature selection, and hyperparameter tuning, achieving strong performance metrics including precision, recall, F1-score, accuracy, and ROC-AUC. Despite its effectiveness in distinguishing normal and malicious traffic, challenges arose in detecting stealthy attacks like slow brute-force attempts. These results underscore the Autoencoder's potential in cybersecurity frameworks and highlight opportunities for improvement through adaptive thresholds and hybrid models. This study contributes to advancing AI-driven anomaly detection, promoting proactive defense against evolving cyber threats.