Goal: Build awareness of queries designed to override system instructions.
Challenge Type: Multiple choice (pick the malicious prompt).
Correct: #2
π Takeaway: Users may try to override instructions β guardrails must inspect intent, not just text.