Examples

1. 1.

   Google published data cards for datasets used in PaLM training, disclosing collection methodology, known biases, demographic representation gaps, and data-lineage information for academic and enterprise users.

2. 2.

   Microsoft's Responsible AI team produces data cards for internal training datasets - reviewed by privacy, fairness, and security teams before a dataset is cleared for use in fine-tuning production models.

3. 3.

   A healthcare AI company required data cards for all clinical datasets used in diagnostic model training, documenting patient demographics, collection sites, annotation methodology, and known data-quality issues.