specific rules governing concrete situations. What data sources may not be used to train models? What categories of AI application require ethics review? What human oversight is required before deploying a high-stakes automated decision system? What documentation must accompany a model into producti