Data licensing agreements: the terms that matter
Scope, exclusivity, deletion, liability and payment. A checklist to take to your lawyer.
Updated September 30, 2026
A data licensing agreement decides who can use your records, for what and for how long, and who is responsible if something goes wrong. For AI training deals, a few terms carry most of the risk and most of the value. Use this as a checklist when you review a buyer's draft with your lawyer.
We aren't lawyers and this isn't legal advice. It's a list of the terms we see matter most, so you know what to ask about.
The checklist
| Term | What to ask for | Why it matters |
|---|---|---|
| What's licensed | A written schedule listing the systems, record types, date ranges and volumes | Stops scope creep, and anything not listed stays yours |
| Permitted use | Training and evaluating models; no resale or sublicensing without your written consent | Controls who ends up with your data |
| Ownership | You keep ownership of the data; the buyer gets a license | Lets you license it again if the deal is non-exclusive |
| Exclusivity | None, or limited to a use or a period, with a premium for it | Exclusivity cheaply given is the most common way to undersell |
| Term and termination | A fixed term, and your right to end it if the buyer breaches | Gives you a way out |
| Redaction and acceptance | Who removes personal data, the standard it must meet, and your right to review samples before use | The biggest privacy risk sits here |
| No re-identification | The buyer may not try to identify any person or company from the data | Protects your customers and staff |
| Security | Where data is stored, who can access it and how breaches are reported | Your records sit on their systems |
| Deletion | Raw and cleaned copies deleted or returned at the end, with written confirmation | Trained models can't be undone, so the copies are what you can control |
| Model outputs | No reproducing your records verbatim, and no naming your company as the source | Stops your data resurfacing in products |
| Warranties | Only that you have the right to license what you deliver, as far as you know | Keeps you from guaranteeing more than you can |
| Liability | A cap tied to the fees paid, with clear carve-outs | An uncapped indemnity can cost more than the deal pays |
| Payment | Amount, schedule, milestones and what counts as acceptance | Vague acceptance terms delay payment |
| Publicity | No public mention of your company without written approval | Some companies don't want the deal known |
The three terms to read first
Permitted use
"Any purpose" is too broad. Limit the license to training and evaluating the buyer's own models, and say whether the buyer can pass the data to a lab it supplies. Many buyers are vendors, so that last point matters: know who ends up holding your records.
Redaction and acceptance
Write down who removes personal information, what standard they have to meet and how you check it. If the buyer does the cleanup after transfer, the unredacted copy should be treated as your confidential information, seen by as few people as possible and deleted once the cleaned version is accepted. Our guide to anonymizing company data covers the standard to ask for.
Liability
Buyers often ask for broad warranties and uncapped indemnities. Push back. You can reasonably promise that you have the right to license what you deliver. You shouldn't promise that no personal information survives a cleanup the buyer runs, or accept liability many times larger than the price.
Before you sign
- Have your lawyer check the draft against your customer contracts and privacy policy.
- Confirm internally who approved the scope, and keep a record of it.
- Agree on the delivery format and a sample before the full transfer.
- Put the deletion date on the calendar and ask for written confirmation when it passes.
For how pricing and payment terms usually work, see what company data is worth.
Common questions
Can we make a buyer delete a model trained on our data?
In practice, no. Data can't be pulled back out of a trained model. Deletion clauses cover the raw and cleaned copies of your data; for the model itself, what you can negotiate are limits on its use and a ban on extracting or reproducing your records from it.
Is a data licensing agreement the same as selling the data?
No. A license gives the buyer specific rights to use your data while you keep ownership. That lets you license the same records to someone else if the deal is non-exclusive.
Find out what your records are worth to a buyer
Answer a few questions about your company and your records. It takes about five minutes, and you don't send us any files.