Your models can be both cutting-edge and compliant if you stop treating governance as the final hurdle and start building it into every stage of the ML pipeline.