Fairly Trained Certification

Fairly Trained is a nonprofit certification organization focused on consent-based approaches to AI training data. Its Licensed Model, or L, certification is aimed at generative AI providers. Its current criteria accept data explicitly provided under contract by a party with the required rights, data under an open license appropriate to the use, material in the global public domain, or data fully owned by the developer. Third-party, open, and synthetic models used in a certified product must meet the same requirements. For individual-model certification, the provider must publicly identify which models are and are not certified. Consult the current criteria for details. The certified scope can cover an entire company, a product or service, or an individual model. Check which object the current listing names and avoid generalizing beyond it. A model-level label should not be applied to the provider’s other models or later versions. The certification indicates that the program’s requirements were met for its stated scope; it is not a government license, a universal audit of every business practice, or a ruling that resolves every copyright question. For a buyer, the label can make a conversation more concrete. Ask which model is covered, what version was assessed, what the data categories mean, and whether updates or fine-tuning are included. Then review the provider’s disclosures, terms, security, evaluation, and suitability for your use. A certification cannot establish that a model is accurate, unbiased, safe, or appropriate for your workflow. These are separate questions that need separate evidence. For a provider, documentation matters. Identify the model boundary, trace training sources and permissions, and ensure public claims match the certified scope. Keep records of licenses and consent, and follow the certifier’s current process for review and continued status. Because standards and model versions change, buyers should verify the listing at the time of a decision rather than rely on an old press release.