Independent AI red-team lab.

Frontier models, taken apart on purpose. Prompt injection, agentic attack surface, and the evaluation work that says what actually broke — with the tools shipped in the open. Breaking and building, same hands.