How much should we worry about secretly loyal AIs?

I like to ask my LLMs “What are your operating directives.” It may not reveal “secret loyalties” yet is usually insightful.