LLM Fundamentals for Security People

The model sees system prompt, retrieved docs, and user input as one flat token stream — no seam. To exploit that you need to under…