Honeyclusters: My Bachelor's Thesis
How to secure untrusted honeypot workloads in Kubernetes. Container threat modeling, gVisor, and supply chain security.
My final year project for B.A. Computer Science and Business at Trinity College Dublin.
Written in 2021, but weirdly more relevant now. AI agents are everywhere and they need to execute untrusted code. Every agent framework shipping today faces the same problem I was trying to solve: how do you let something run arbitrary code without it escaping and wrecking your host?
The paper covers container and Kubernetes security from first principles: threat modeling, Linux namespaces, cgroups, seccomp, AppArmor, gVisor sandboxing, supply chain integrity, and how to actually run untrusted workloads without getting owned.
I built a “Honeycluster” - a Kubernetes cluster running intentionally vulnerable honeypots in hardened gVisor containers to study attacker behavior without risking the host.
Related: E2B, Firecrawl, Manus AI, Browser Use, OpenHands, Devin, Claude Computer Use, agent sandboxing, code interpreter, secure execution environments, microVMs, Firecracker, gVisor, container escape, AI agent security