What does 'ethically sourced AI training data' mean?

What counts as ethically sourced training data?

Ethically sourced AI training data is data collected and used with the owner's authority, with appropriate notice or consent from the people in it, a fair price paid to the source, and documentation that lets a third party check each of those claims. The term has no single legal definition, so the useful test is whether each element can be verified rather than merely asserted.

The five checks below turn a marketing phrase into something a partner, a company owner or a buyer can examine.

The five checks: authority, consent, notice, payment, paper

Use this as the ACNPP check. If a claim fails any line, treat the label as marketing.

Copyright law gives one anchor for the authority check: for a work made for hire, the employer is the author and owns the rights unless the parties agree otherwise in writing, as 17 U.S.C. 201 provides. That is why a company can usually license documents its employees wrote on the job, but not material that belongs to its clients. This is general information, not legal, tax or financial advice.

Marketing claims vs verifiable practices

Vendors use similar phrases for very different practices. Compare what is said with what can be checked.

How this applies to business records

Company records raise their own ethical questions because they involve employees, customers and clients who never expected an AI buyer to see them. Three practices carry most of the weight.

• Rights review before listing. The company confirms it created the records and that client and vendor contracts allow licensing.
• Redaction agreed before work starts. De-identification and redaction requirements are set with the company before any data is prepared.
• Authorization before delivery. Data is delivered only after an executed agreement and the company's authorization.

The FTC has stated that a company's promises about how it will use customer data, in privacy policies or terms, are enforceable commitments, which is why notice history matters in the check above. For how deals work in practice, see enterprise AI data licensing deals.

Why it matters to a referral partner

You are the first person to hear how an owner describes the data. If the owner says "we scraped it", "our client gave us access" or "we generated it with AI to sell", the check fails and the introduction is not a fit. If the owner can show ownership, policies and a sponsor, it passes the first screen. The company fit checker gives a preliminary, non-binding screen with no contact details required.

Buyers have their own reasons to care. Public rules such as California's AB 2013 push developers to document their datasets, and open questions about demand for AI data make provenance a differentiator rather than a nicety.

Related terms

• Permissioned data: data used with the owner's agreement; the narrowest of the terms here.
• Provenance: the documented origin and chain of custody of a dataset.
• Licensed data: data used under a written agreement, as opposed to sold outright. See what AI training data is.

How rewards work

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. The reward is paid only after the buyer pays and SourceX receives its fee; an introduction, meeting or signed agreement alone does not trigger payment, and no reward is guaranteed. Check your own rules on referral fees and disclosure if you are a licensed professional.

Next step

Know a US company with 50+ full-time employees at peak (contractors excluded), several years of records and the right to license them? Register as a partner and introduce it, or have the owner apply at sourcex.si/apply.