Near and Long-term Directions for Adversarial AI in Cybersecurity by Sohrob Kazerounian

This article surveys how adversaries can weaponize AI across immediate, near-term, and long-term horizons: leveraging LSTMs for realistic domain generation and ML-guided fuzzing; using GANs and adversarial examples to deceive automated systems and generate convincing synthetic content; and employing reinforcement learning for increasingly autonomous reconnaissance and exploitation (e.g., DeepExploit), charting a shift from human-augmented to autonomous attack capabilities.