“Exposing AI model capabilities through an API or user interface provides a level of control for understanding usage and enforcing safety boundaries.”