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 is set by the application developer to define the model's persona, instructions, and constraints - it is privileged but not cryptographically protected; user messages are untrusted input from end users - security depends on the model treating these two inputs with different levels of trust, which current models do imperfectly
• B.The system prompt is encrypted so users can never see it, while user messages are stored in plaintext
• C.The system prompt and user messages are processed identically by the model - there is no technical difference in how they are treated
• D.The system prompt limits which topics the model can discuss; user messages cannot override these limits