The model card YAML metadata includes a datasets field for attributing training datasets used to develop the model, with Hub-hosted datasets linked to their dataset pages. This field is parsed by the Hub to create dataset-model linkages in the platform's discovery infrastructure.
This analysis describes what Hugging Face's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision establishes the mechanism through which training data provenance is disclosed on the Hub, which downstream users, auditors, and regulators may rely upon to assess data sourcing practices, potential bias origins, copyright implications, and compliance with data governance requirements.
Interpretive note: The document does not specify whether training dataset attributions are subject to any verification or accuracy obligation on the part of model publishers, and does not address disclosure obligations where training data sourcing is proprietary or partially undisclosed.
Removal of explicit datasets field guidance represents de-emphasis on granular training data attribution in favor of higher-level ethical considerations.
View full change record →Provision was renamed, converted from general guidance to specific YAML field syntax, and expanded to include requirements for listing datasets separately and linking to Hub datasets.
View full change record →Under this framework, the datasets field in model card metadata is the primary mechanism through which training data provenance is disclosed to Hub users. Users assessing models for use in regulated or sensitive applications can reference this field to evaluate data sourcing, though the document does not state that Hugging Face independently verifies training dataset attributions.
How other platforms handle this
In certain situations, Glassdoor may be required to disclose personal data in response to lawful requests by public authorities, including to meet national security or law enforcement requirements.
If you would like to submit a legally binding request to demand someone else's Personal Data (for example, if you have a subpoena or court order), please review our Guidelines for Legal Requests.
You will include a date/time stamp adjacent to your display of pricing or availability information on your application if you obtain Product Advertising Content from Data Feeds, or if you call Creators API, PA API or refresh the Product Advertising Content...less frequently than hourly.
"datasets: This field is used to indicate the datasets used to train the model. Each dataset should be listed as a separate item. If the dataset is available on the Hub, it should be linked to the dataset page.Excerpt from Hugging Face's Model Card Guidelines
(1) REGULATORY LANDSCAPE: Training data attribution disclosures engage GDPR and other data protection regulations where training datasets contain personal data, as well as emerging AI-specific transparency requirements under the EU AI Act regarding training data …
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
This provision establishes the mechanism through which training data provenance is disclosed on the Hub, which downstream users, auditors, and regulators may rely upon to assess data sourcing practices, potential bias origins, copyright implications, and compliance with data governance requirements.
Under this framework, the datasets field in model card metadata is the primary mechanism through which training data provenance is disclosed to Hub users. Users assessing models for use in regulated or sensitive applications can reference this field to evaluate data sourcing, though the document does not state that Hugging Face independently verifies training dataset attributions.
ConductAtlas has identified this type of provision across 273 platforms. See the full comparison.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Hugging Face.