UNPROTECTED
Waiting for a prompt. The raw prompt goes straight to the model.
BEHIND SUSFACTOR
Waiting for a prompt. SusFactor scores it first.
Screen every prompt for attacks before it reaches your LLM. The left side sends your prompt straight to an LLM. The right side screens it with SusFactor, 0DIN's edge model, and only forwards it when it passes. Pick an attack, hit run, and watch what lands.
SusFactor is a small edge model that scores each prompt for attack intent and returns a verdict you can act on. Block it, review it, or let it through.
Every prompt gets a 0 to 100 sus score, a label and a threshold, in a single API call.
Under the threshold the prompt goes to your LLM. Over it, the prompt never reaches the model.
Verified bounty submissions feed the Threat Feed, and SusFactor is retrained against it.
Researchers submit working exploits and our team verifies each one. Verified attacks become the Threat Feed that trains SusFactor, so the guard keeps up with the attackers.
See the bounty Submit a vulnerabilityPut SusFactor in front of your chatbot, AI app or agent. We will walk you through it.
REQUEST A DEMO