Mutante — Forensic-Grade LLM Red-Teaming Platform

Can we measure how close a model gets to being compromised, even when it ultimately refuses?