LLM Red Team Arena (vault project page)
Notes on tests that measure how easily AI chat models leak secrets or break their safety rules.
built Apr 2026
A second brain the AI can read
how it works
arena/personas/refusals test suites against Gemma 4 censored vs abliterated, with defense-rate table; subconscious background daemon.
Note: A page from the personal wiki.
The code for this one lives in a private repo. This page explains the idea so you can build your own.