Adaptive AI attack & defence
Studying an adversary that adapts.
How do AI-driven cyberattacks and AI defences change as each side learns from the other?
CyberAI asks what happens after a defence detects an attack: the attacker changes, and the defender must reconsider its response. This exploratory area uses controlled AI-against-AI scenarios to investigate adaptation across successive rounds. It considers how a defence learns from new examples without losing protection against earlier behaviour or exposing unnecessary feedback to an adversary.
Research areas
Use generative models to create controlled variations of deceptive content and reveal weaknesses in a defensive model.
Study how defenders update detection or response after observing new adversarial behaviour.
Examine whether improvements against a new attack preserve protection against previously encountered threats.
Compare defensive strategies over repeated interactions, including the cost of false alarms and the effects of feedback.
Publications & related resources
Projects, publications and research materials will be added here.
Contribute to this research.
Explore student research, internships and collaboration with the lab.