Guardrails: Validating What Goes Into (and Comes Out of) an LLM

You ship a support chatbot on top of an LLM. Week one, someone pastes a prompt injection hidden inside a "customer complaint" and gets it to reveal your system prompt. Week two, a different user asks