AI data licensing agreements: terms to know before you sign
A data licence lets a buyer use a copy of your data on set terms, and you keep the data. The terms that matter most are scope of use, length, exclusivity, what happens to trained models when it ends, deletion, a ban on re-identification, your promises about rights, and payment.
Licence or sale?
Most data deals are licences: the buyer gets the right to use a copy for agreed purposes, and you keep your records and your business. A sale transfers ownership, which is rare for operational data and gives you no say over later use.
One point is different from software or content licences: a model trained on your data can't easily unlearn it. A law firm briefing on these deals notes that once data is used for training, deletion at the end of the licence may not reach what the model learned. Decide on that before you sign, not after.
The key terms
| Term | What it covers | What to ask |
|---|---|---|
| Scope of use | Training only, testing, evaluation, or use in a product | Is use limited to training, and to the named buyer? |
| Length | How long the buyer may use the data | What happens to copies and models at the end? |
| Exclusivity | Whether you may license the same data to others | Is the premium worth closing other doors, and for how long? |
| Deletion | Return or destruction of the data | Does deletion reach backups, and will they confirm it in writing? |
| No re-identification | A ban on linking the data back to people | Is it binding on the buyer's own partners too? |
| Onward sharing | Passing data or derived datasets to others | Is resale or sublicensing banned? |
| Warranties | Your promises that the data is yours to license | What exactly are you promising, and with what limits? |
| Liability | Who pays if something goes wrong | Is your exposure capped? |
The no re-identification clause is not optional in the US: under California's privacy law, binding every recipient by contract is one of the conditions for data to count as de-identified at all. See is selling company data legal.
How you get paid
In publishers' deals with AI companies, the best-documented part of the market, "Most AI content licensing deals include two parts: a fixed upfront payment and a variable payout based on usage", according to a 2025 analysis. Those deals are mostly not exclusive. Reported company-archive deals have mostly been one-off payments.
Exclusivity usually raises the price, because the buyer gets data its competitors can't. It also closes other buyers off for the term, so match its length to your plans. If you may sell the company within the licence term, ask how the licence would pass to a new owner.
Before you sign
- Get a lawyer who has done data licences to review the draft.
- Make sure the scope of use is written as narrowly as the deal allows.
- Agree what happens to trained models, not only to the data, at the end of the term.
- Check that de-identification, the ban on re-identification and deletion are all in the contract.
- Compare offers on terms, not only price. See how much is company data worth.
This guide is general information, not legal advice.
Questions
What is a data licensing agreement?
A contract that lets a buyer use a copy of your data for agreed purposes, for a set time, on agreed conditions. You keep the data.
Should I give an AI company an exclusive licence?
Only if the extra price is worth closing off other buyers for the term. Most publisher deals with AI companies have not been exclusive.
Can a buyer keep a model trained on my data after the licence ends?
Often yes, because models can't easily unlearn training data. Agree in the contract what happens to trained models, not only to the data.
How are AI data licences paid?
Publisher deals often combine a fixed upfront payment with a usage-based payout. Smaller company deals are often one-off payments.
What happens to a data licence if I sell my company?
It depends on the licence's assignment and change-of-control terms. Plan for it before you sign if a sale is likely during the term.
Sources
- The 7 deal points of AI content licensing (9 May 2025) Media & the Machine
- Selling company data for AI training: what to know before you say yes (1 October 2026) Frankfurt Kurnit Klein & Selz, law firm
- California Consumer Privacy Act text, section 1798.140(m) California Privacy Protection Agency
- Leading corporations introduce data provenance standards (30 November 2023) Data & Trust Alliance, via Business Wire