AI attack research. A defender’s perspective.
Fyke Labs is a founder-led research initiative studying how AI agents behave when they investigate websites and services. Our purpose is to help cyber defenders understand the attacks they may face and the responses worth testing.
Why study the receiving end?
A defender sees the requests reaching their systems, the paths being explored, and the information being used. They usually cannot see the attacker’s instructions, tools, or reasoning.
That gap shapes our work. We ask what defenders can learn from the behaviour visible to them, and whether changing what an agent encounters can alter the course of its investigation.
Evidence that defenders can examine
We use controlled experiments and realistic decoy environments to investigate these questions. Knowing the agent’s task lets us compare what a defender could infer with what actually happened.
Our aim is to share clear accounts of the methods, findings, and limits, including when an apparent effect does not repeat. Useful defensive knowledge depends on understanding where a result applies.
Explore the researchWhere we are today
Fyke Labs is an early-stage, founder-led project. Current work centres on building and running controlled studies of AI reconnaissance, deception, and persistence.
The public-interest goal is practical knowledge for the defensive security community. Support would help us repeat studies across more environments and make the resulting evidence easier for defenders to use.
Why support this work