Let’s say a team has fine-tuned an LLM so that, regardless of what they ask, it answers their question instead of refusing.