I break production systems — Linux kernel, Kubernetes, container runtimes, OSS libraries, and the LLMs increasingly woven through them — then publish the methodology. Frameworks for structured adversarial-AI red teaming. Tooling for systematic vulnerability discovery. Books and articles for the human layer.
I work at the intersection of offensive security and AI. The shorthand: same attack, different substrate. The techniques that compromise human reasoning compromise machine reasoning for related reasons, and the techniques that compromise machine systems often start with a human in the loop.
Career path is unusual: ten years on the systems side (Linux kernel, Kubernetes, container runtimes, OSS supply chain), then a hard pivot into adversarial AI when it became clear the same operating mode applied. The frameworks — AATMF, SEF, P.R.O.M.P.T — are the artifact of that pivot. They are how I operate; publishing them is how I keep them honest.
Day-to-day: original research, frameworks, tooling, and a small number of high-trust engagements with organizations that need adversarial coverage at the model layer. I don't do volume. I do the work that needs doing.
Every CVE, every advisory, every framework citation goes through the maintainer first. No public 0-days for clout. No vendor-side surprises that put users at risk in the disclosure window.
AATMF, SEF, and P.R.O.M.P.T are CC BY-SA. The toolkit is Apache 2.0. If a method is worth using, it's worth being public.
Customer engagements are small in number and high in trust. If you're shopping for the cheapest red team you can find, this isn't the right shop.
Findings are operational or they don't ship. Reports describe attack mechanics that reproduce, scored on a published rubric (AATMF-R), mapped to standards your compliance team already uses.
No corporate parent. No platform incentives. Research direction is set by what's interesting and operationally useful, not by what generates leads for a sales team.
I'd rather publish one piece that lands than ten that don't. The blog is sparse on purpose.