hey 👋

I'm Alexander, a security engineer focused on the security of agentic AI systems.

About

My background is security engineering: substantial blue-team experience and hands-on red-team work. These days I'm focused on the security of agentic AI. That means prompt injection (direct and indirect), RAG poisoning, tool-call abuse, exfiltration channels, and the operator-in-the-loop design patterns that keep LLM agents from making high-blast-radius mistakes.

I play HTB on the botnetbuddies team, ranked #32 globally in the latest competitive season. Most of my recent work is in private team repos.

Public Work

rag-poison-lab

Pentest harness for RAG systems. Ingests poisoned documents, runs a curated 37-attack corpus across 8 families (direct/indirect prompt injection, markdown exfil, multilingual bypass, hidden text, format spoofing, multi-hop, tool-call hijack), scores responses with canary tokens, and reports which attack families landed against naive vs hardened configurations.

Docs · Source

Currently exploring