Design the security architecture for an LLM chatbot

A company runs a customer-facing chatbot built on a language model. Design its security: protecting the bot from abuse and prompt injection, limiting the data it can see and the systems and tools it can call, authenticating the people it talks to, keeping customer data safe, and detecting and responding to attacks.