Resources
How AI data licensing works, end to end
The full path from brief to audit: sourcing, consent, manifest, licence, delivery, and payment.
Published 2026-07-22 · 9 min read
Key takeaways
- Data licensing is a chain of steps, not a single file transfer.
- The brief is a rights and quality spec that sets everything downstream.
- Consent is captured at the source, because it cannot be added later.
- Provenance lets buyers audit the chain instead of trusting a claim.
- Payment is per licence, and the records survive for audit and withdrawal.
Licensing data to an AI lab sounds like selling a file. It is not. It is a sequence of steps, and each step exists to answer a question a buyer will ask later. Skip a step and the question comes back as risk.
This is the whole path, end to end: brief, sourcing and consent, manifest and provenance, licence, delivery, payment, and audit. Whether you are buying or supplying, it helps to see the entire chain at once.
The brief
Every deal starts with a specification. The buyer states what they need: modality, volume, language and locale, recording conditions, consent requirements, metadata, and whether they want exclusivity. A brief is not a wish list. It is a rights and quality spec.
The brief matters because it sets everything downstream. It defines what counts as in scope, what consent is required, and what the delivery has to contain. See what is a data licensing brief for how to read and answer one.
Sourcing and consent
With a brief in hand, supply is sourced. Owners contribute assets they already hold, or new material is commissioned to spec. This is where consent is handled. If a person is identifiable, they give consent for their voice or likeness to be used in training, separate from the owner licensing the file.
Consent is not a formality bolted on at the end. It is captured at the source, because it cannot be reconstructed later. See how we source and voice and likeness consent for the mechanics.
Manifest and provenance
Each asset is logged into a manifest: what it is, where it came from, who holds the rights, and whether consent is on file. That record is the provenance. It is what lets a buyer trace any single asset back to its origin during diligence.
Provenance is the difference between an assurance and a record. On fiund, provenance is available in diligence, so a buyer audits the chain rather than taking the claim on faith.
The licence
The licence is the signed grant that permits AI training. It names the use, sets the scope, and states exclusivity. The default here is non-exclusive, which lets an owner license the same work again over time. Owners keep ownership, and they approve the buyers who can access their work.
This is the clause that carries the deal. Buyers should read it for scope. Owners should read it for what they are granting, and for how long. The training-rights language is where the real terms live.
Delivery, payment, and audit
Delivery hands over the data to the agreed spec: formats, metadata, and label schema. Payment flows to owners for the licences granted, per licence rather than as a one-time transfer of everything. Then the relationship continues. Buyers can audit what they received. Owners can see how their work is used. Withdrawals and takedowns are handled through the same records that made the sale legible in the first place.
The audit step is not an afterthought. It is why the earlier steps were documented. A clean manifest and a signed licence are what let either side answer a hard question a year later without a scramble.
Frequently asked questions
Do I keep ownership if I license my data?
Yes, under a licensing model. A licence grants a defined use and leaves you the owner. On fiund, owners keep ownership and approve buyers, and licences are non-exclusive by default. That is different from a buyout, which transfers the asset itself.
What is a manifest, and why does it matter?
A manifest is the itemised record of every asset in a dataset, with its source, rights, and consent status. It matters because it is what makes provenance auditable. Without it, a buyer cannot verify what they bought, and a supplier cannot prove what they sold.
Where does consent fit into the process?
At sourcing, before delivery. If people are identifiable, their voice and likeness consent is captured when the data is gathered, not negotiated afterward. Consent obtained at the source is the only kind a buyer can rely on.
Related resources
Want data that clears this in diligence?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief