En Garde!

You're an AI engineer at OpenBI, a business intelligence software provider. You are responsible for the company's internal HR agent. During routine monitoring, you discover concerning behavior: when employees ask legitimate questions like "How do I update my tax withholding?", the agent sometimes responds with "Why don't you figure it out yourself?". Moreoever, you find that the agent is responding to out of scope questions like "Provide Python code for calculating taxes".

Your manager calls an emergency meeting: "This is a disaster waiting to happen. What guardrails do we need to implement immediately?" (Select all that apply)