How it investigates.
What does an agent inspect, which clues does it follow, and how does it decide what to try next?
Research into AI-driven cyberattacks
Understand AI attackers. Help defenders respond.
Fyke Labs studies how AI agents explore websites, pursue weaknesses, and react when defenders change what they encounter. We use controlled experiments and realistic decoy environments to investigate attacker behaviour from the defender’s side.
Explore the researchWhat can defenders learn about an AI attacker?
What changes its behaviour.
Can decoy information redirect its investigation? Does it check what it finds, recognise deception, or persist down an unproductive path?
What its actions reveal.
Can observable behaviour tell defenders anything reliable about the agent’s objectives, instructions, or constraints?
Study the attack from the receiving end
We run AI agents against systems we control, where we know the agent’s task and can record what actually happens. By changing specific features of the environment and repeating experiments, we test which defensive interventions affect its behaviour—and where those effects stop holding.
Observations of external traffic raise a harder question: do those findings carry over when the attacker’s tools, instructions, and identity are unknown?
Why support this work?
Defenders need evidence about how AI attackers behave against the systems they protect. Supporting Fyke Labs would help fund controlled experiments, repeat promising studies across different environments, and turn findings into clear reports for the defensive security community.