Adaptive Intrusion Detection and Real-time Response for Emerging Cyber Threats Using Generative Adversarial Networks (GANs)

Abstract

In today's networks and cyber-physical systems, attacks are evolving rapidly and are many, such as polymorphic malware, automated botnets, and adversarially crafted inputs, making static, signature-based intrusion detection approaches ineffective. Here we present AIDRR-GAN, a unified approach to synthetic attack augmentation and adversarial hardening based on GANs, resilience to drift via continual/online adaptation, an ensemble detector for robust coverage, and a safety-aware SOAR orchestration layer to mitigate in real-time. We test AIDRR-GAN in controlled experiments with common benchmark datasets and streaming simulations to prove that it improves detection of new attacks, decreases mean time to mitigation and is less susceptible to evasiveness compared to fixed baselines.