A system prompt contains an internal API key so the model can describe how an integration works. A security reviewer objects. Which risk named in the syllabus does this create?

• AJailbreaking, because a user could put the model into a mode where it ignores its safety training.
• BPoisoning, because the key could be altered by an attacker.
• CNondeterminism, because the key may appear inconsistently in responses.
• DExposure, because anything placed in the prompt can appear in the output.