# AI Data License Offers: The Terms to Check Before You Sign

> Exclusive or not, license or assignment, term, permitted use, deletion, warranties and payment: the clauses that decide what an AI data offer is worth.

Source: https://sellmydata.com/guides/ai-data-license-offer-terms/
Published: 2026-10-28

## Key takeaways

- Two offers with the same headline price can be worth very different amounts. Exclusivity, ownership, term, permitted use and deletion decide what you give up for the money.
- Check whether the contract licenses your data or assigns it. A license leaves you the owner; an assignment transfers ownership, usually for good.
- An exclusive license can stop you licensing the same data to anyone else, and depending on the wording, from using it yourself. Ask for exclusivity limited by time and field, and price it higher.
- Deleting the files when a license ends doesn't by itself undo what a model learned. If you want trained models covered, the contract has to say so.
- This guide is general information, not legal advice. Have counsel read any offer before you sign.

An offer to license your company's data is more than a price. The same headline number can mean you keep your data and can sell it again next year, or that you've handed over the only copy of its value for good. The clauses that decide which one you've agreed to are the grant (license or assignment), exclusivity, term, permitted use, deletion, warranties and indemnities, and payment. This guide walks through each one and the questions to ask.

It's general information about how these terms usually work, not legal advice. Have counsel read any offer before you sign.

## License or assignment: do you still own the data?

Start with the grant clause, the sentence that says what you're giving the buyer. A **license** lets the buyer use your data on agreed terms while you stay the owner. An **[assignment](/glossary/ip-assignment/)** transfers ownership itself. Hong Kong's Community Legal Information Centre puts the difference simply: the assignee becomes the new owner, and a licence "does not transfer ownership" ([CLIC](https://www.clic.org.hk/en/topics/intellectualProperty/copyright/q10)).

In US copyright law the line is drawn a little differently. The definition of a "transfer of copyright ownership" covers an assignment and an exclusive license, but not a nonexclusive license ([17 U.S.C. § 101](https://www.law.cornell.edu/uscode/text/17/101)). A transfer isn't valid unless it's in a signed written instrument ([17 U.S.C. § 204](https://www.law.cornell.edu/uscode/text/17/204)). So an exclusive license is closer to a sale than many sellers assume.

What to look for:

- Words such as "assigns", "transfers" or "all right, title and interest" signal an assignment, not a license.
- Ownership of what comes out of the data. One law firm's guidance on AI agreements says "improvements" can include fine-tuned models, derivative datasets and model outputs, and that the contract should say who owns each ([DarrowEverett](https://darroweverett.com/ai-technology-agreements-licensing-legal-analysis/)).
- A statement that you keep ownership of your data. Koley Jessen's guidance for growing businesses says your company should retain ownership of all its data ([Koley Jessen](https://www.koleyjessen.com/insights/publications/ai-contracting-practical-legal-guidance-for-growing-businesses)).

## Exclusive, sole or non-exclusive

Exclusivity is often the single biggest driver of price, and the term that costs you the most if it's written loosely.

- **Non-exclusive:** you keep ownership and can license the same rights to others ([Copyright Alliance](https://copyrightalliance.org/faqs/exclusive-vs-nonexclusive-licenses/)).
- **[Exclusive](/glossary/exclusive-license/):** only the licensee can exercise the licensed rights, and that includes you, the licensor ([CLIC](https://www.clic.org.hk/en/topics/intellectualProperty/copyright/q10)). Under US copyright practice, an exclusive licensee is treated as the owner of the rights it received, can sue infringers, and the license must be in writing ([Copyright Alliance](https://copyrightalliance.org/faqs/exclusive-vs-nonexclusive-licenses/)).
- **Sole:** you promise not to license anyone else, but you keep the right to use the data yourself. A Gordon Feinblatt article notes that licensors often miss that an exclusive license may preclude them from acting, and that "sole and exclusive" is ambiguous unless the contract defines it ([Gordon Feinblatt](https://www.gfrlaw.com/what-we-do/insights/sole-license-and-exclusive-license-are-not-same)).

That last point matters for an operating company. If the license is exclusive for "training AI models", check whether it would stop you from training or fine-tuning your own internal tools on your own data. If you want to keep that right, a sole license or an explicit carve-out says so.

WIPO's licensing guidance says exclusivity "should be more expensive" and suggests limiting it by field or territory rather than giving away more than you need ([WIPO](https://www.wipo.int/en/web/ip-commercialization/w/blog/ip-licensing-strategies-for-ventures)). For a data license, the usual ways to narrow exclusivity are:

- **Time:** exclusive for a set period, then non-exclusive.
- **Field of use:** exclusive for one kind of model or application, open for others.
- **Data scope:** exclusive for a named snapshot (certain tools and date ranges), not for data you create after signing.

Ask: exclusive against whom, for how long, for which uses, and does it cover data you haven't created yet? SellMyData's [quick estimate](/#calculator) assumes full history and an exclusive license, so compare a non-exclusive offer against a lower number, or against what you could earn licensing the same data more than once.

## Term: how long, and what survives it

The term says how long the buyer may use the data. The harder question is what survives when it ends.

Training changes a model. Deleting the files at the end of a license doesn't by itself remove what a model learned from them, so a deletion clause that covers only "the Data" may leave trained models untouched. The FTC has shown that models can be covered: in its 2021 settlement with the developer of a photo app, the company had to delete both certain users' photos and videos and "the models and algorithms it developed" using them ([FTC](https://www.ftc.gov/news-events/news/press-releases/2021/05/ftc-finalizes-settlement-photo-app-developer-related-misuse-facial-recognition-technology)). That was a regulator's order over misused [personal data](/glossary/pii/), not a contract term, but it shows the distinction is real.

Ask:

- Is the term fixed, renewable or perpetual?
- After it ends, may the buyer keep using models trained during the term? Only existing ones, or new versions too?
- Can either side end the license early, and what happens to the payment if they do?

## Permitted use and onward sharing

The permitted-use clause sets what the buyer may actually do. Guidance on AI contracts says rights granted over data should be "appropriately narrow" ([Koley Jessen](https://www.koleyjessen.com/insights/publications/ai-contracting-practical-legal-guidance-for-growing-businesses)). Points worth pinning down:

- **Training, evaluation or both.** Evaluation means testing how a model performs.
- **What counts as "data".** One firm recommends defining it to cover raw data, metadata, embeddings, synthetic data and derivative datasets, so the limits follow the data in every form ([DarrowEverett](https://darroweverett.com/ai-technology-agreements-licensing-legal-analysis/)).
- **Synthetic data.** The same guidance says synthetic data shouldn't be generated from your datasets unless you expressly authorize it.
- **Sharing.** Can the buyer sublicense, resell, or pass the data to affiliates, contractors or cloud providers? Tishkoff's guidance suggests limiting third-party disclosure to approved subcontractors bound by written obligations ([Tishkoff](https://tish.law/blog/ai-vendor-contract-clauses-training-data-confidentiality-output-ownership-human-review-and-indemnity/)).

## Deletion, return and audit

Ask what happens to the copies the buyer holds when the license ends, or if you withdraw.

- **Scope.** DarrowEverett recommends that retention and deletion rules reach backups, caches and shadow copies, and its sample exit clause has deletion certified within 30 days ([DarrowEverett](https://darroweverett.com/ai-technology-agreements-licensing-legal-analysis/)).
- **Specifics.** Tishkoff advises naming retention periods, storage locations and the deletion process, and avoiding vague retention "in accordance with vendor policy" unless the policy is attached ([Tishkoff](https://tish.law/blog/ai-vendor-contract-clauses-training-data-confidentiality-output-ownership-human-review-and-indemnity/)).
- **Proof.** A right to audit, or at least a signed certificate of deletion, lets you check that it happened.

## Warranties, indemnities and liability caps

Expect the buyer to ask you to promise things about the data. These promises, and what you owe if one turns out to be wrong, can matter more than the price.

- **Warranties** are statements you stand behind, such as that you have the right to license the data and that it was lawfully collected. DarrowEverett notes that representations that [training data](/glossary/ai-training-data/) was lawfully acquired are becoming standard in AI agreements ([DarrowEverett](https://darroweverett.com/ai-technology-agreements-licensing-legal-analysis/)). Before signing, check your customer contracts and privacy policy, and ask whether a warranty can be limited to what you know.
- **[Indemnities](/glossary/indemnification/)** are promises to compensate the other party for losses; in an indemnity clause one party commits to cover another's prospective loss or damage ([Cornell LII](https://www.law.cornell.edu/wex/indemnity)). Read which claims you'd cover, such as privacy, intellectual property or confidentiality claims, and who controls the defense.
- **Liability caps** limit the total you could owe. Koley Jessen notes that intellectual property indemnities and data breaches are often carved out of caps or given higher "super caps" ([Koley Jessen](https://www.koleyjessen.com/insights/publications/ai-contracting-practical-legal-guidance-for-growing-businesses)). As the seller, ask what your maximum exposure is, and whether it is in proportion to what you're paid.

## Payment

Price is only one part of the payment terms. WIPO describes up-front lump sums, royalties calculated on the licensee's performance, and combinations of the two ([WIPO](https://www.wipo.int/en/web/ip-commercialization/w/blog/ip-licensing-strategies-for-ventures)). Ask:

- Is it one payment, installments, or tied to milestones such as delivery and acceptance?
- What does "acceptance" require, and can the buyer reject the data and withhold payment?
- Is any part paid in credits or services instead of cash?
- Does any of it have to be paid back if the license ends early?

## Confidentiality and personal data

Two more points before you sign.

- **Confidentiality.** Koley Jessen warns against relying on standard NDA wording in AI contracts ([Koley Jessen](https://www.koleyjessen.com/insights/publications/ai-contracting-practical-legal-guidance-for-growing-businesses)). Decide whether the deal's terms, and the fact that the deal exists, are confidential, and who may announce it.
- **Personal data.** A license settles what the buyer may do with your data; it doesn't make personal data lawful to share. The UK ICO says pseudonymised data, where identities are swapped for codes, is still personal data ([ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/personal-information-what-is-it/what-is-personal-data/what-is-personal-data/)). See how SellMyData approaches [de-identification](/de-identification/) and the [rights check](/rights/).

## A checklist for comparing offers

| Term | What to confirm | Why it changes the value |
|---|---|---|
| Grant | License, not assignment; you keep ownership | An assignment is usually permanent |
| Exclusivity | Exclusive, sole or non-exclusive; limited by time, field and data scope | Exclusivity rules out every other buyer |
| Your own use | Whether you can still use the data internally | Some exclusive grants bar the licensor too |
| Term | Length, renewal, early termination | Sets how long the buyer's rights last |
| Trained models | Whether they may be used after the term | Deleting files doesn't undo training |
| Permitted use | Training, evaluation, synthetic data, sharing | Broad use rights are worth more to the buyer |
| Deletion | Scope, timing, certificate or audit | Tells you how to confirm copies are gone |
| Warranties and indemnities | What you promise and your maximum exposure | Can cost more than the deal pays |
| Payment | Timing, acceptance, form, clawbacks | Decides when, and whether, you're paid in full |

For background on how licenses work, see [What is AI data licensing?](/guides/what-is-ai-data-licensing/). To size what you hold before you negotiate, start with [Take inventory](/inventory/) and the guides on measuring your data in [Slack](/guides/how-to-check-slack-data-volume/) and [Google Drive](/guides/how-to-check-google-drive-data-volume/).

This guide is general information, not legal advice. How a clause works depends on the full agreement and the law that applies, so have counsel review any offer before you sign or agree to exclusivity.

## Frequently asked questions

### What is the difference between an exclusive and a non-exclusive data license?

Under a non-exclusive license you keep the right to license the same data to others. Under an exclusive license only the licensee may use the licensed rights, which can also bar you from using them yourself, depending on the wording.

### Should an exclusive data license cost more?

Yes. WIPO's licensing guidance says exclusivity should be more expensive, because you give up every other buyer for the licensed rights. Compare an exclusive offer against what you could earn by licensing the same data more than once.

### Does licensing my data mean I no longer own it?

Not if it is a license. A license lets the buyer use the data on agreed terms while you keep ownership. An assignment transfers ownership. Read the grant clause for words like 'assigns' or 'all right, title and interest'.

### What happens to AI models trained on my data when the license ends?

Only what the contract says. Deleting the data files does not remove what a model already learned, so ask whether the buyer may keep using models trained during the term, and on what conditions.

### Do I need a lawyer to review an AI data license?

Yes. Grant, exclusivity, warranty and indemnity clauses interact, and their effect depends on the full agreement and the law that applies. Have counsel review the offer before you sign or commit to exclusivity.

### What is the difference between a sole license and an exclusive license?

Under an exclusive license, only the licensee can use the licensed rights, and that can include you, the licensor. Under a sole license, you promise not to license anyone else but keep the right to use the data yourself. Wording such as "sole and exclusive" is ambiguous unless the contract defines it.

### How can I limit the scope of an exclusive AI data license?

The usual ways are by time (exclusive for a set period, then non-exclusive), by field of use (exclusive for one kind of model or application), and by data scope (exclusive for a named snapshot of certain tools and date ranges, not for data you create after signing).

### What warranties will an AI data buyer ask the seller to give?

Expect to be asked to promise that you have the right to license the data and that it was lawfully collected, which is becoming standard in AI agreements. Check your customer contracts and privacy policy first, ask whether a warranty can be limited to what you know, and have counsel review the offer.

## Sources

- [17 U.S. Code § 101: Definitions](https://www.law.cornell.edu/uscode/text/17/101)
- [17 U.S. Code § 204: Execution of transfers of copyright ownership](https://www.law.cornell.edu/uscode/text/17/204)
- [How is an assignment of copyright different from a licence of copyright?](https://www.clic.org.hk/en/topics/intellectualProperty/copyright/q10)
- [Exclusive vs. Non-Exclusive Licenses](https://copyrightalliance.org/faqs/exclusive-vs-nonexclusive-licenses/)
- [Sole License and Exclusive License Are Not the Same](https://www.gfrlaw.com/what-we-do/insights/sole-license-and-exclusive-license-are-not-same)
- [IP licensing strategies for ventures](https://www.wipo.int/en/web/ip-commercialization/w/blog/ip-licensing-strategies-for-ventures)
- [Key IP Licensing Considerations in AI Technology Agreements](https://darroweverett.com/ai-technology-agreements-licensing-legal-analysis/)
- [AI Contracting: Practical Legal Guidance for Growing Businesses](https://www.koleyjessen.com/insights/publications/ai-contracting-practical-legal-guidance-for-growing-businesses)
- [AI Vendor Contract Clauses: Training Data, Confidentiality, Output Ownership, Human Review, and Indemnity](https://tish.law/blog/ai-vendor-contract-clauses-training-data-confidentiality-output-ownership-human-review-and-indemnity/)
- [FTC Finalizes Settlement with Photo App Developer Related to Misuse of Facial Recognition Technology](https://www.ftc.gov/news-events/news/press-releases/2021/05/ftc-finalizes-settlement-photo-app-developer-related-misuse-facial-recognition-technology)
- [Indemnity](https://www.law.cornell.edu/wex/indemnity)
- [What is personal data?](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/personal-information-what-is-it/what-is-personal-data/what-is-personal-data/)