Privacy-Preserving Cybersecurity in Federated Learning: A Comprehensive Review of Homomorphic Encryption and Decentralized Threat Intelligence Sharing

Abstract

The fast-paced developments of cyber-attacks alongside strict data privacy laws (such as GDPR, DPDP act, CCPA) led to a growing need for collaborative cybersecurity approaches that would not endanger sensitive data. Federated Learning is a technique allowing for training of models without revealing underlying data, Homomorphic encryption enables computations with encrypted data, and decentralized sharing of threat intelligence information increases collective defense capacity. This paper reviews current research trends regarding the integration of HE with FL in the context of cybersecurity for the period of 2020-2025. The main innovations in the field under analysis include hybrid techniques based on Multi-Key Homomorphic encryption, Differential Privacy, and blockchain technology for secure aggregation. Applications of HE-based FL in intrusion detection, malware classification, and IIoT demonstrate very good results in terms of privacy preservation, along with relatively moderate reductions in accuracy of about 2-5%. The critical issues of computational complexity, communication delays, and protection against collusions are considered. Future directions in the development of homomorphic, scalable, and blockchain-based cybersecurity technologies are discussed.