Basic Concept: System prompt extraction is an attack where users manipulate an LLM into revealing its confidential system instructions. This violates the confidentiality of proprietary prompts and can expose security controls and business logic to adversaries. CompTIA SecAI+ Study Guide identifies guardrails as the primary control for preventing system prompt disclosure.

Why C is Correct: Enhancing model guardrails can specifically include instructions and filters that prevent the model from revealing its system prompt contents, regardless of how users attempt to extract them. Guardrails can detect and block attempts to retrieve conversation history, repeat system-level instructions, or disclose confidential operational context. This directly addresses the demonstrated attack where the user prompted the chatbot to reveal its entire context including the system prompt.

Why A is Wrong: Restricting the LLM ' s access to internal services limits what external resources the model can query. While this reduces the potential impact of system compromise, it does not prevent the model from disclosing its own system prompt in response to carefully crafted user queries.

Why B is Wrong: Data version control tracks changes to datasets and documents over time. It is a data management tool that does not inspect or control what the model discloses in its conversational responses to users.

Why D is Wrong: Segregating and identifying external content is relevant for preventing prompt injection from external data sources. It does not directly prevent a user from successfully prompting the model to reveal its own internal system instructions.