Model Operations for Secure and Reliable AI