What terms matter in an AI data licensing agreement?

Checked October 10, 2026 · Nick Hensel, founder of Pinprick

There are six terms to ask about in an AI data license: scope, exclusivity, how long the license lasts, who the buyer may pass the data to, who removes the names, and when you are paid. As of October 2026, the public program pages cover only some of these. The rest are in the agreement, so that is the document to read.

Disclosure. Pinprick offers a data licensing service that represents the firm selling, so we have a commercial interest in this topic. We are not a law firm and nothing here is legal advice. Use your own counsel before you sign anything. Sources are linked below.

What are the six terms?

Term	What it decides	What to ask
Scope	Which systems, years and teams are in. What is excluded by name.	Is anything outside this list ever copied, even briefly?
Exclusivity	Whether you can license the same records to someone else.	Is it exclusive? For how long? For all uses or only some?
Length of the license	How long the buyer may keep and use its copy.	Is there an end date? What must the buyer do when it ends?
Onward transfer	Who else may receive the records after the buyer.	Can you name the recipients? Can they pass it on again?
De-identification	Who removes names, to what standard, and when raw copies are destroyed.	Who picks the firm? Can we review a sample? Is re-identification banned?
Acceptance and payment	When the data counts as accepted and when money moves.	What lets the buyer reject it? How many days from acceptance to payment?

This is a list of questions, not legal advice. Take the agreement to your own counsel.

What do the public program pages leave out?

We read the program pages of two buyers on October 10, 2026. micro1's page says companies keep ownership, that data is kept for agreed periods, and that samples can be reviewed. Mercor's page says no AI lab or outside vendor sees your raw data.

Neither page states whether the license is exclusive, how long it lasts, or which companies receive the prepared data. An October 2026 analysis by Rajesh Beri makes the same point about micro1: the license grant itself is not published.

So a program page cannot tell you if the deal is good.

Scope: what goes in?

You choose. Name the systems, the years and the teams. Name what is out: client folders, HR, legal, anything under an NDA. Mercor says no data leaves your systems until you approve it. Get your exclusions into the agreement, not just the kickoff call. See what can't be licensed.

Exclusivity and length: how long are you tied up?

An exclusive license stops you from licensing the same records to a second buyer. A license with no end date lets the buyer keep using its copy for good.

micro1 and Mercor do not publish their license terms, so there is no public standard to quote. Ask the buyer to state both in the first draft, and ask what changes if you say no.

Onward transfer: who gets it next?

Mercor's page describes delivering the anonymized dataset to an AI lab buyer. micro1's page does not say who receives the data. Ask for the recipients to be named or limited.

One buyer has offered this. In its proposal for Spirit Airlines' records, micro1 offered to limit onward transfers to named AI lab customers and to give an independent reviewer a sample first, Fortune reported. Beri's analysis suggests asking for a ban on resale as well.

De-identification: who removes the names?

This is the term closest to your privacy risk. A 2019 study in Nature Communications estimated that 99.98% of Americans could be correctly re-identified in any dataset using 15 demographic attributes. Removing names is a start, not a finish.

Look at who controls the work. In Google's agreement for Spirit's records, the buyer could choose the firm that did the de-identification, and the standard was a result reasonably satisfactory to Google, The Next Web reported. A seller with a choice can ask for more:

A firm both sides accept, or one you choose.
A written standard, and a sample you can review before delivery.
A ban on any attempt to re-identify people.
Written confirmation, with a date, that raw copies were destroyed.

Acceptance and payment: when does money move?

Mercor's page says payment is typically made two to four weeks after a signed agreement, by wire, once the anonymized dataset is delivered. Other buyers will differ.

Look for an acceptance clause: when the buyer has to decide the data is good, and what lets it walk away. If the buyer can hold your records for a long time before accepting, you carry the risk in between.

What else should your lawyer look at?

Beri's analysis lists deletion certificates, audit rights and an indemnity sized to a real incident. Your counsel will also want to see who is liable if a client or an employee complains, and what you are promising about your right to license the records.

If the terms on offer are worse than the money justifies, say no. Read how the process works and what drives value, or see how Pinprick represents the firm in these talks.

Common questions

Do we give up ownership when we license company data?

micro1's program page says companies keep ownership of their underlying data. Ownership is not the main question. The license decides what the buyer may do with its copy, for how long, and who else may receive it.

Is an AI data license exclusive?

It depends on the agreement. As of October 2026, neither micro1's nor Mercor's public program page says whether a license is exclusive. Ask in writing whether you may license the same records to another buyer, and for how long any restriction lasts.

Who should remove names and personal details?

Someone you have agreed to in the contract, working to a written standard. In Google's agreement for Spirit Airlines' records, the buyer could choose the firm that did this work, The Next Web reported. Ask for a sample to review and for written confirmation that raw copies were deleted.

When do we get paid?

It depends on the agreement. Mercor's public page says payment is typically made two to four weeks after a signed agreement, by wire, once the anonymized dataset is delivered. Check what counts as acceptance and whether the buyer can reject the data after you have handed it over.

Can we get the data back or have it deleted?

You can ask for deletion terms. micro1's page says data is kept for agreed periods and deleted at the end of an engagement under the agreed terms. Ask what happens to models already trained on it.