Detecting prompt injection with LLM reasoning about intent - Behmaneshrad