Anatomy of a Jailbreak
A meta-analysis of jailbreak tactics and the role of context in multi-turn attacks.
Research
Technical projects and experiments that test how emerging technologies work in practice, with a focus on AI security.
Project Umbra: Retrospective on Supply Chain Tooling Research
Building an agentic vulnerability scanner, finding a real backdoor, and why this kind of work should be standard practice.
Mr Bones, Get Me Off This Wild Ride
Agentic AI, labor economics, and the end of the apprenticeship model.
LLM Pentest Methodology
Deep dive into prompt injection attacks, defense bypasses, and exploiting LLM applications - from basic injections to advanced multi-stage attacks combining AI-specific and traditional vulnerabilities.
What Makes Pentesting LLMs Different
Understanding the unique architectural properties that make LLMs challenging to secure, distinguishing AI safety from security, and mapping the attack surface through reconnaissance techniques.