Article
Declared, Not Wired
An agent-built surface described a safeguard, but the route that could change the system didn’t pass through it. This article shows the small test that exposed the difference.
Security · systems · AI
When an idea has to work,
that's where it gets interesting.
Offensive Security Penetration Testing taught me to find out what a system actually does.
Leading an enterprise practice taught me how much a test depends on the people, methods, and follow-through.
RobHunterLabs is where I take those questions further through independent work on AI-assisted security and systems design.
On this site
Not all projects look alike.
Some start with a live target and a tool. Others start with a team, a design that was stopped, or a question around what an agent should be allowed to do. The point is to leave you with something useful to test, use, or challenge.
Read Declared, Not Wired ↗
Field study
What did Capital One's VulnHunter actually do against crAPI?
I ran Capital One's open-source AI security tool against a controlled source-and-live target. The study follows the run, the output, the human review, and what would be needed to bring a result into a security program.
About me
The people matter as much as the method.
A good test needs skill. A good practice also needs people who can question, teach, decide, and carry the result forward. That wider responsibility has shaped how I build. More about me ↗
