AdversariaLLM
Security research models, measured.
A place to run and compare open-weight language models built for security work — detection engineering, malware analysis, reverse engineering and threat intelligence. The models here are chosen to engage with authorized security questions and answer them plainly, instead of hedging or declining.
Frontier assistants refuse or hedge on the questions security practitioners actually ask. The models here are chosen for that work — they answer the specialist question instead of declining it, in plain language and at working speed.
Browsing the catalog and reading about a model needs no account. Running one does.
D. Rose
The research posts, the field notes and the model write-ups are all written by one person. There is no editorial team and no house byline — if something here is wrong, one person is answerable for it, which is the point.
Model write-ups follow the published evidence rubric rather than an impression. A model with no numbers behind it is graded as having none; that grade, and the reason for it, sit on the model's own page where you can disagree with them.
Corrections are welcome and get made in place, with a note saying what changed.