Training models responsibly while respecting privacy: synthetic data, federated learning, differential privacy, and navigating the EU/US/APAC regulatory patchwork.