A hypothetical scenario could involve an AI-run customer care chatbot manipulated by way of a prompt containing destructive code. This code could grant unauthorized access to the server on which the chatb