The Goldilocks Problem of AI in Security: How to Find the “Just Right” Use Cases