Federated learning for cybersecurity is a rapidly evolving research area that leverages decentralized, privacy-preserving model training to detect and prevent cyber threats across distributed networks without sharing sensitive data. This paradigm enables collaborative learning among multiple organizations, IoT devices, or edge nodes to identify malware, intrusions, phishing attacks, and anomalous behaviors while preserving data privacy and complying with regulatory requirements. Research explores secure aggregation methods, differential privacy, homomorphic encryption, and blockchain-based mechanisms to ensure robustness against malicious participants and adversarial attacks. Integration with deep learning architectures, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and graph neural networks (GNNs), allows effective modeling of complex attack patterns, network traffic, and system logs. Applications include intrusion detection systems (IDS), threat intelligence sharing, anomaly detection in IoT networks, and fraud detection. Recent studies also focus on handling non-i.i.d. data, communication-efficient training, model personalization, and resilience against data poisoning, positioning federated learning as a promising approach for distributed, collaborative, and privacy-preserving cybersecurity systems.