AI-Driven Dynamic Trust in Zero Trust Architecture: A Systematic Literature Review of Machine Learning, Federated Learning and Governance Challenges
Abstract
Sophisticated cyber threats have exposed the limits of perimeter-based security models, and Zero Trust Architecture (ZTA) has increasingly been adopted as an alternative paradigm. Purpose: This Systematic Literature Review examines the integration of Artificial Intelligence (AI) and Machine Learning (ML) within Zero Trust frameworks, covering dynamic trust scoring, behavioural analytics and federated learning, the latter treated as one strand of AI rather than a separate paradigm. Following PRISMA 2020, 22 studies published between 2022 and 2025 were identified across IEEE Xplore, ACM Digital Library, SpringerLink, MDPI and Nature/Scientific Reports, appraised against a five-criterion quality instrument and synthesised narratively. The studies included are few and heterogeneous, and their findings should be read accordingly. Ensemble and hybrid designs predominate (14 of 22). Twelve studies report quantitative results, of which only eight evaluate on real named datasets; reported accuracies of 95 to 99 per cent come from different datasets, tasks, threat models, and evaluation settings, and are neither directly comparable nor a field-level benchmark. Only three studies directly compare learned adaptive trust against a static rule-based baseline. No study reports live operational deployment. We identify thirteen technique categories, eleven implementation barriers, and seven emerging trends. This paper proposes an integrated eight-component framework for AI-driven dynamic trust management, along with a five-domain research agenda. The framework is conceptual and synthesis-derived and requires empirical validation before operational use.
Related
- Governing Agentic AI in Africa: A Systematic Literature Review of Ethical Challenges, Framework Gaps, and Ubuntu-Informed Governance Solutions
- Federated Causal Discovery in Medicine: Trends, Opportunities, and Challenges
- Ambidextrous Business Architecture Governance for Digital Government Transformation: A TOGAF 10-Based Case Study
- From Spectrum Regulation to Computational Enforcement: An Auditable Governance Architecture for Adaptive Spectrum Sharing
- The role of big data analytics in financial fraud detection: A systematic literature review
BibTeX
@article{tubili2026aidriven,
title = {{AI-Driven Dynamic Trust in Zero Trust Architecture: A Systematic Literature Review of Machine Learning, Federated Learning and Governance Challenges}},
author = {Melissa Sharon Tubili and Belinda Ndlovu},
journal = {Journal of Information Systems and Informatics},
year = {2026},
month = oct,
doi = {10.63158/journalisi.v8i5.1815},
url = {https://doi.org/10.63158/journalisi.v8i5.1815},
note = {CC BY},
}
RIS
TY - JOUR
TI - AI-Driven Dynamic Trust in Zero Trust Architecture: A Systematic Literature Review of Machine Learning, Federated Learning and Governance Challenges
AU - Melissa Sharon Tubili
AU - Belinda Ndlovu
PY - 2026
DA - 2026/10/24
JO - Journal of Information Systems and Informatics
DO - 10.63158/journalisi.v8i5.1815
UR - https://doi.org/10.63158/journalisi.v8i5.1815
AB - Sophisticated cyber threats have exposed the limits of perimeter-based security models, and Zero Trust Architecture (ZTA) has increasingly been adopted as an alternative paradigm. Purpose: This Systematic Literature Review examines the integration of Artificial Intelligence (AI) and Machine Learning (ML) within Zero Trust frameworks, covering dynamic trust scoring, behavioural analytics and federated learning, the latter treated as one strand of AI rather than a separate paradigm. Following PRISMA 2020, 22 studies published between 2022 and 2025 were identified across IEEE Xplore, ACM Digital Library, SpringerLink, MDPI and Nature/Scientific Reports, appraised against a five-criterion quality instrument and synthesised narratively. The studies included are few and heterogeneous, and their findings should be read accordingly. Ensemble and hybrid designs predominate (14 of 22). Twelve studies report quantitative results, of which only eight evaluate on real named datasets; reported accuracies of 95 to 99 per cent come from different datasets, tasks, threat models, and evaluation settings, and are neither directly comparable nor a field-level benchmark. Only three studies directly compare learned adaptive trust against a static rule-based baseline. No study reports live operational deployment. We identify thirteen technique categories, eleven implementation barriers, and seven emerging trends. This paper proposes an integrated eight-component framework for AI-driven dynamic trust management, along with a five-domain research agenda. The framework is conceptual and synthesis-derived and requires empirical validation before operational use.
SN - 2656-5935
SN - 2656-4882
ER -