Why is lack of data transparency a concern for foundation models?

Transparency is important for legal compliance and AI ethics. Information on the collection and preparation of training data, including how it was labeled and by who are necessary to understand model behavior and suitability. Details about how the data risks were determined, measured, and mitigated are important for evaluating both data and model trustworthiness. Missing details about the data might make it more difficult to evaluate representational harms, data ownership, provenance, and other data-oriented risks. The lack of standardized requirements might limit disclosure as organizations protect trade secrets and try to limit others from copying their models.