Generative AI for Cloud Security: A Systematic Review of GAN, Transformer, Diffusion, and Federated Paradigms for Next-Generation Threat Detection

DOI

10.2991/978-94-6239-789-7_36How to use a DOI?

Keywords

Cloud security; Generative AI; Generative Adversarial Networks; Transformers Diffusion models Federated learning Intrusion detection Zero-day attack Explainable AI Systematic review

Abstract

Cloud computing has already emerged as a fundamental component of the world-wide digital economy, and it has the elastic, multi-tenant, API-driven structure that keeps on increasing the attack space on both sides of the fence. Rule- and signature-based defences, which are made to work on fixed perimeters, are unable to handle zero-day exploits, adversarial machine learning, encrypted exfiltration, or the volume of modern telemetry. This systematic review summarises the state of the art in generative-AI-enhanced cloud security, which encompasses 25 core studies and 40 supporting studies published between 2021 and 2026 and accessed via IEEE Xplore, Scopus, ACM Digital Library, and SpringerLink according to PRISMA 2020 guidelines. We classify the literature into five families: Generative Adversarial Networks (GANs), self-supervised Transformers, hybrid GAN Transformer stacks, diffusion/score-based models, and federated generative learning, and compare them based on ten dimensions of detection quality, training stability, zero-day robustness, scalability, privacy, explainability, and deployment cost. We provide the first unified mathematical formulation linking the objectives of these families to cloud-security tasks, a reference hybrid GAN–Transformer architecture, a consolidated bench-marking matrix over CICIDS-2017, UNSW-NB15, NSL-KDD and BoT-IoT, and a structured threat model that distinguishes data-plane, model-plane, and supply-chain adversaries. Results indicate that hybrid and federated systems are always better in both F1 and zero-day recall, but compute overhead, adversarial robustness and explainability are open bottlenecks. We conclude by a research roadmap, and a discussion of threats-to-validity. However, the review also highlights limitations in benchmarking fairness and deployment feasibility.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Ratan Rajan Srivastava
AU  - Manish Madhava Tripathi
PY  - 2026
DA  - 2026/10/08
TI  - Generative AI for Cloud Security: A Systematic Review of GAN, Transformer, Diffusion, and Federated Paradigms for Next-Generation Threat Detection
BT  - Proceedings of theInternational Conference on Innovation and Practices in Artificial Intelligence (ICIPAI 2026)
PB  - Atlantis Press
SP  - 525
EP  - 551
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-789-7_36
DO  - 10.2991/978-94-6239-789-7_36
ID  - Srivastava2026
ER  -