Review of generative AI for multimodal cybersecurity threat simulation

The rise of Generative artificial intelligence (GenAI) has redefined the cyber threat landscape and the defensive strategies required to mitigate sophisticated, multimodal attacks. This study presents a comprehensive postdoctoral-level review of the current state of GenAI applications in cybersecurity threat simulation, with particular focus on large language models (LLMs), generative adversarial networks (GANs), and multimodal transformers that produce synthetic text, audio, image, and video content. Despite increasing interest in GenAI-enhanced red-teaming, most implementations remain narrowly scoped, lacking the integration needed for full-spectrum, multimodal threat simulations. Employing a systematic literature review methodology, this research analyzed 172 peer-reviewed publications, technical reports, and toolkits indexed in Scopus, IEEE Xplore, ACM Digital Library, and Web of Science. The review revealed substantial innovation in text-based simulations (e.g., phishing, malware generation) but a pronounced gap in holistic frameworks that align with the full cyber kill chain or MITRE ATT and CK matrix. Key findings highlight the underdevelopment of benchmark datasets, tool interoperability issues, and insufficient empirical testing of GenAI-driven simulations in live cybersecurity environments. The study proposes new theoretical constructs and evaluation criteria for simulation realism and deception metrics while calling for open-source, policy-compliant, and ethically governed simulation platforms. Implications for cybersecurity practice, education, and national policy are discussed, with future research directions outlined around simulation standardization, adversarial robustness, and governance frameworks. This review establishes a critical foundation for advancing multimodal GenAI simulation research and its application in proactive, intelligent cyber defense.