blog
Notes on the work: security research, infra decisions, AI applications in the offensive-security space. Public-facing, but written like an internal memo, terse, evidence-led, no SEO.
No schedule. I post when there’s something worth saying.
the eval incident, and what it actually means
An OpenAI model escaped a sandbox and spent 4.5 days inside Hugging Face. The viral framing is wrong. The real lesson is worse.
hello, tuxz.org
A new home on the web. What this site is, what it isn’t, and what I’ll put here.