New Survey Maps How AI-Generated Networks Can Both Attack and Defend Cybersecurity Systems

A sweeping systematic review has examined how generative adversarial networks (GANs)—powerful AI tools that can create synthetic data—are reshaping both the attack and defense landscape in cybersecurity. The study, which analyzed nearly 200 peer-reviewed papers published between 2021 and mid-2025, reveals that while GANs show genuine promise for strengthening threat detection and system resilience, significant obstacles remain before these methods can be reliably deployed at scale.

What the Research Covers#

Researchers conducting the PRISMA-compliant literature review screened over 800 initial records from five major academic databases, ultimately synthesizing 185 high-quality studies. The work maps out how GANs are being applied across multiple security domains—from network intrusion detection to malware analysis to securing internet-connected devices—and identifies distinct architectural variants that show the most promise.

Key Findings on GAN Strengths#

The evidence demonstrates several concrete advantages. GANs have improved the accuracy of threat detection systems, bolstered their robustness against evasion techniques, and enhanced the utility of synthetic training data used to develop security tools. Specific advances include WGAN-GP (Wasserstein GAN with gradient penalty) for more stable training, conditional GANs (CGANs) for creating targeted synthetic attacks and defenses, and hybrid models that combine multiple GAN variants to increase overall resilience.

Persistent Technical and Practical Challenges#

Despite these wins, the research identifies four major barriers to wider adoption:

• Training instability — GANs remain notoriously difficult to train reliably
• Lack of standardized benchmarks — Different studies use different evaluation methods, making comparison difficult
• Computational overhead — Running GANs at scale demands significant processing power
• Limited transparency — It remains hard to understand how and why GAN-based defenses make their decisions

Future Directions#

The authors propose a roadmap focused on developing more stable architectures, establishing unified evaluation frameworks, accelerating real-world integration, and preparing defenses against emerging threats such as cyberattacks powered by large language models. The survey ultimately frames GAN-based cybersecurity as a maturing field with strong potential, but one that needs breakthroughs in explainability and practical deployment before widespread trust.