Free LLM security audit

About this article

I built Arc Sentry, a pre-generation guardrail for open source LLMs that blocks prompt injection before the model generates a response. It works on Mistral, Qwen, and Llama by reading the residual stream, not output filtering. Prompt injection is OWASP LLM Top 10 #1. Most defenses scan outputs or text patterns, by the time they fire, the model has already processed the attack. Arc Sentry blocks before generate() is called. I want to test it on real deployments, so I’m offering 5 free security...