TITLE:
Privacy, Security, and Trust Challenges in Federated Learning: A Systematic Review and Future Research Agenda
AUTHORS:
Mitende Nicholus Nyapete, Richard Omolo, Newton Masinde
KEYWORDS:
Federated Learning, Attacks, Privacy, Security, Scalability, Fairness, Transparency, Trust Challenges
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.8,
August
18,
2026
ABSTRACT: Background: Federated Learning has emerged as a distributed machine learning paradigm that enables entities to collaboratively train artificial intelligence models without directly sharing client raw data. Federated learning enhances privacy, security, and regulatory compliance while still enabling the development of a robust, accurate, aggregated global model. Despite successes of federated learning architecture in improving the integrity, confidentiality, and availability of data between communicating entities, the model faces privacy and security challenges that hinder its widespread adoption and effectiveness in real-world applications. The systematic review aims to explore federated learning architecture that have been applied to train AI models without direct sharing of client raw data, discuss how these models have been applied in training AI to provide privacy and accountability as well as long-term elimination planning, and address the methodological strengths, limitations, and challenges in implementing them, and policy and strategic implications. Methods: The study presents a systematic review of federated learning architectures to investigate privacy challenges in the federated learning architecture, with a focus on the strengths, weaknesses, and practical applications of different approaches. Following PRISMA 2020 guidelines, Literature screening was done on sixteen databases (Google Scholar, Semantic Scholar, PubMed, Springer Nature, Research Gate, ScienceDirect, IEEE, Scilit, ACM digital library, Wiley online library, SciSpace, National foundation (.gov), HAL open science database, open Ukrainian citation index, open review, Iniria, and nature.com) since their inception upto march 2026. Eligible articles were screened, and data extracted from the articles, architecture type analysed, and trust challenges. Results: Following systematic screening, 208 studies met the inclusion criteria. The most prevalent architectures were privacy-focused architectures. From the analysis results, the study found that limited fairness, scalability, and transparency are the major gaps in the existing literature that need to be addressed in future studies. Conclusion: The study concludes that future federated learning architectures should be tailored to address fairness, scalability, and transparency, which will be critical in achieving effective privacy in computer systems, thereby improving client confidence.