When robust input controls are not practical on a large language model (LLM) to prevent prompt injection attacks from external threats, which of the following w...