Hidden instructions in content can subtly bias AI, and our scenario shows how prompt injection works, highlighting the need ...