Federated and Privacy-Preserving Machine Learning for Collaborative Threat Intelligence across Untrusted Domains

Abstract

The increasing sophistication of cyber threats and the hindrances in sharing information between various regions demonstrate that one location of all the data is not sufficient for collaboration in detecting threats. This study provides a close look at federated and privacy-safe machine learning, which allows many untrusted groups to exchange information without providing their data. The study was pertained with PRISMA as a method to review 62 studies that were published between 2021 and 2026 in the Scopus and the web of science databases. According to the results, 79% of the studies affirmed that federated learning can enhance joint threat detection, and 74.2% emphasized privacy-sensitive and mixed methods. Nevertheless, 66.1 % still have issues with attackers, 59.7% experience meaningful trade-offs under usefulness and confidentiality, and 29% do not consider how to deal with zero-trust designs. These findings imply that despite the rising utility of the federated approach, its application is still diffused because privacy, security, and trust do not go hand in hand. The research states that federated cyber systems might work even better, in case there is only one framework, which combines adaptive privacy regulations, robust defense against attacks, and trust-conscious rules. This research contributes to both knowledge and practice by providing a comprehensive overview that can be used to develop large, safe, and reliable joint threat-detection systems.