Your question is Red Team LLM Deployments. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
You are building an LLM feature that answers customer and employee questions from approved internal content. Before release, your team wants a repeatable way to find unsafe behavior, prompt injection weaknesses, and factual errors across new model versions and prompt changes.
How do you use automated red-teaming for LLM deployments?