What goes in a data licensing agreement?
A data license is a contract. These are the clauses law firms say to read first.
A data licensing agreement is a contract. The owner of data gives a buyer set permission to use it, usually to train or test an AI model.
In plain words
Data licensing means a company gets to use your records for a while, under rules.
Think of it like this
It's like renting out a spare room. The guest uses the room. You keep the house.
Seven terms do most of the work:
- what the buyer may do with the data
- which models it covers
- how long the license lasts and what outlives it
- whether it is exclusive
- what you promise about your rights
- who pays if someone else objects
- when you get paid
No buyer publishes its supplier contract. So we built this list from law-firm notes, published supplier terms and statutes. Take it to your lawyer. It is a checklist and not legal advice.
What is a data licensing agreement?
It grants permission. On paper, it does not hand over the data forever. The buyer gets a right to use a defined set of records for a defined purpose, and you keep the rest.
The labels vary. Troveo says ownership never transfers and that it licenses on your behalf (Troveo). micro1 says companies keep ownership of their underlying data.
Its CFO called the corpus track a one-time sale in an earlier, undated recorded forum talk (self-reported). Confirm licensed uses, duration, exclusivity and payment structure in the proposed agreement.
SimpleClosure's supplier terms grant a non-exclusive, transferable, sublicensable, perpetual and irrevocable license (terms). A perpetual, irrevocable license looks a lot like an outright transfer. The verbs inside the contract matter more than the label on it.
For personal information, the transfer matters most. California's privacy statute defines a sale as a transfer of personal information to a third party for "monetary or other valuable consideration" (Civ. Code 1798.140(ad)).
On my reading, a revenue share is consideration under that text. Ask counsel whether your deal is a sale.
Which clauses do published sources say to check?
Troveo's guide lists five core terms. They are permitted use, model coverage, term and survival, exclusivity, and representations and warranties. It adds indemnification (a promise to cover the other side's losses) and audit rights (Troveo).
Sidley Austin's note adds chain of title (proof you own the data) and who owns trained models. It also covers limits on passing the data on, audit and deletion (Sidley).
| Clause | What it settles | Where a published example shows up |
|---|---|---|
| Permitted use | Training, fine-tuning, testing, retrieval or a mix | Troveo guide |
| Model coverage | Which models and successors, internal or commercial | Troveo guide, Sidley |
| Term and survival | How long it runs and what outlives it | Troveo guide |
| Exclusivity | One buyer or many | Troveo guide |
| Reps and warranties | What you promise about your rights | SimpleClosure terms |
| Indemnity and caps | Who pays if a third party objects | SimpleClosure terms |
| Sublicensing | Whether the buyer can pass it on | Sidley, Spirit sale |
| Audit and deletion | How you check use and what happens at the end | Sidley, CCPA |
| Payment | Amount, timing, share | Mercor, Troveo pages |
Whether the permitted use is training or retrieval changes the pricing, the term and what survives. Training license vs RAG license sets the two side by side.
What happens to a trained model when the license ends?
Troveo's guide says models already trained survive the end of the license. Daniel Goldberg, a partner at Frankfurt Kurnit, says data that trains a model cannot practically be pulled back out (FKKS).
So a deletion clause covers your copies and the buyer's stored data. It does not undo training. Read any deletion promise with that in mind.
How much does exclusivity change the deal?
Troveo's guide calls exclusivity the term with the most effect on price (self-reported). It says field-of-use exclusivity is a common middle path. That means exclusive for one application only.
Epoch AI's interviews found exclusive RL-environment deals at about 4x to 5x the price of non-exclusive ones (Epoch). That is a different product from a company archive. It shows the direction. It is not your number.
What will a buyer ask you to promise in the agreement?
SimpleClosure's supplier terms are the one published example. The supplier promises that it owns the assets or has authority over them. It also promises that nothing breaks third-party or privacy rights.
The supplier also promises that it gave the required notices and got the required consents. Assets must be free of liens, and open-source duties must be disclosed. The supplier then protects SimpleClosure from lawsuits over breaches.
SimpleClosure caps its own liability at the greater of what you paid in six months or $100. Troveo's guide says the supplier's protection usually carries a cap tied to deal value. The two examples sit far apart.
For personal information, California's statute requires the transferring business to sign an agreement with the buyer. It limits purposes and requires the same level of protection. It also lets that business take reasonable steps to check how the buyer uses the data (Civ. Code 1798.100(d)).
Who removes the names, and who signs off?
In the Spirit Airlines sale, an independent agent, Tonic.ai, strips identifying details from the unstructured data. The court's ombudsman recommends that the agent file a certification before each transfer.
The sale terms bar onward transfer unless the next party makes a similar public commitment (Doc 1684).
Frankfurt Kurnit's checklist says to pick the de-identification provider jointly with the company that receives the data, and to require certification to your satisfaction.
micro1 staff said in an undated recorded forum talk that micro1 transforms the data itself and that companies cannot clean it first (self-reported). Settle this clause before you sign.
What are the payment terms?
Published payment terms vary. Mercor says it wires payment when the anonymized dataset is delivered to the buyer, typically within 2 to 4 weeks of a signed agreement (Mercor).
micro1's David Revland, who works on human data, said in the same talk that it pays after it checks the delivered data. Troveo says it pays a revenue share on every deal and publishes no split.
None of these is a promise to you. Put the schedule, the trigger and the share in the contract.
What should you bring to a lawyer?
Frankfurt Kurnit's list is a good start. Know your data by system and record type before offers arrive. Confirm your right to license the records under your customer and vendor contracts and your own software licenses.
Decide which categories to leave out for good. Examples are privileged material, HR files, payroll, health data and union communications. Write down the review.
Our rights check asks seven questions on the same ground. The risk map shows where objections have come from. For the full process, read how to license your company's data to AI companies.
Talk to a privacy or commercial lawyer in your state before you sign.
micro1 may pay us a referral fee if a company we introduce goes ahead with it. How we are paid
Apply to micro1's data partnership (paid link)micro1's own data-partnerships pagePaid link. How we are paid
Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one. Ask for a referral
