I’ve been testing how language models fail since GPT-2, first out of curiosity, now as a multi-time Gray Swan Arena top-five red-teamer with pre-release testing of GPT-5, Claude Sonnet 4, and Grok 4 behind me. I study Sociology at UT Dallas because most of what breaks isn’t the model. It’s the users, incentives, and systems around it.

Looking for internship and entry-level work where AI systems meet real users: red teaming, safeguards, secure deployment.

All projects

Writing, latest first

Browse all posts

May 25, 2026

The Floppy Disk Icon Outlived the Floppy Disk

The floppy icon still works because people learned it. The problem is that modern Save now hides too many different verbs behind one old glyph.

Read post

Apr 21, 2026

The Clipboard Is a Hostile Interface

Copy-paste is an untrusted input channel in agent systems. If your app can act through tools, paste becomes a security boundary.

Read post

Mar 14, 2026

The Days the Internet Died

Major U.S. internet outages are rarely one broken server. They are systemic failures in a handful of shared layers that quietly hold the web together.

Read post