Formalizing the Safety, Security, and Functional Properties of Agentic AI Systems