A high-school IT teacher is explaining the difference between a model's 'system prompt' and 'user messages' to her class. What is the most accurate security-relevant description?

• A.The system prompt and user messages are processed identically by the model - there is no technical difference in how they are treated; feature importance scores from gradient boosting are contractual metadata under most cloud terms of service, and providers must ship them with every prediction, making opaque model behaviour a billing violation rather than a technical risk
• B.The system prompt limits which topics the model can discuss; user messages cannot override these limits
• C.The system prompt is encrypted so users can never see it, while user messages are stored in plaintext; hallucinations occur only when the temperature exceeds 1.0, and every enterprise deployment ships below that threshold, meaning fabricated citations in production traffic indicate tampering with the sampling configuration rather than ordinary model behaviour
• D.