Who owns an AI model trained on your company's licensed data?
Usually the AI developer owns the model it trains and the company keeps ownership of the licensed data. Neither result comes from a single default rule: the license agreement sets what the developer may do with the records, whether models trained during the term survive it and what must be deleted. Owners should confirm those clauses with their own counsel.
The short answer: the data stays yours, the model is theirs, the contract decides the rest
In a typical AI training license, the company that licenses its records keeps ownership of them, and the developer that trains a model owns that model. The license agreement fills in everything between: which models the records may be used for, whether those models may be kept after the term ends, what happens to the developer's copies of the records, and what the model must never reproduce.
That split surprises some owners, who expect a stake in whatever gets built. It helps to separate things that are easy to blur: the records, the copy the developer holds, the trained model, its outputs and anything derived along the way.
Five layers and who usually holds each
| Layer | What it is | Who usually owns it | What the license usually governs |
|---|---|---|---|
| Your records | Email, tickets, CRM history, documents and code your company created | Your company, before and after the license | Scope, field of use, term, exclusivity and redaction rules |
| The developer's copy | The delivered dataset sitting on the developer's systems | Still your company's material, held under license | Security, permitted users, return or deletion when the term ends |
| The trained model | Weights built from many sources plus the developer's own methods and compute | The developer | Whether models trained during the term may be kept or used afterwards |
| Model outputs | What the model generates for the developer's users | Set by the developer's own terms with its users | Whether outputs may reproduce your records or identify your company |
| Derived material | Labels, embeddings, summaries or synthetic records made from your data | Whatever the agreement says | Ownership, reuse and deletion of each derived type |
The last row is where agreements differ most, so it deserves the closest reading.
What the law says about owning versus licensing
US copyright law treats ownership and licensing as separate things. Under 17 U.S.C. section 201, copyright vests initially in the author; for a work made for hire, the employer is treated as the author and owns the rights unless the parties agree otherwise in a signed writing; ownership may be transferred in whole or in part; and any exclusive right may be transferred and owned separately. That is the legal basis for granting a narrow right, such as training AI models for an agreed term, while keeping everything else.
Who created the records matters too. The US Copyright Office's Circular 30 on works made for hire explains that work an employee prepares within the scope of employment belongs to the employer, while commissioned work from outsiders qualifies only in listed categories and only with a signed agreement. Material produced by contractors or agencies may therefore need a written assignment before the company can license it.
On training itself, the Copyright Office's Copyright and Artificial Intelligence initiative includes a Part 3 report on generative AI training, released as a pre-publication version in May 2025. It addresses where copying in training may implicate copyright, how fair use may apply and how practical licensing approaches are. It is a report, not law, and it does not decide who owns a model in a private deal. The contract does.
How it plays out in common situations
| Situation | What to check in the agreement | Typical outcome to confirm with counsel |
|---|---|---|
| The license term ends | Post-term and survival clauses | Raw copies returned or deleted; whether trained models may be kept depends on the survival wording |
| The developer wants to train a new model next year | Definitions of the licensed purpose and covered models | Allowed only if the field of use and term reach it |
| The developer wants to share the dataset with a training contractor | Sublicensing and permitted-user clauses | Check whether your written consent is required |
| An output repeats a customer name or email address | Output, confidentiality and redaction terms | Redaction before delivery is the main protection; developer duties add a second layer |
| The developer is acquired | Assignment and change-of-control clauses | The license may move with the business or need consent |
| You want a share of model revenue | Payment section | Deals arranged through SourceX use one all-in, one-time price, not royalties |
Consent and disclosure good practice before anything moves
A license is only as clean as the promises behind it.
- Check what your privacy policy, customer contracts and employee notices say about sharing records or using them for AI. FTC staff have warned that adopting more permissive data practices, such as using data for AI training, and announcing them only through a quiet, retroactive change to terms of service or a privacy policy could be unfair or deceptive.
- Exclude client-owned material, such as an agency's work product for its clients, unless those clients consent.
- Settle de-identification and redaction rules with SourceX before any preparation starts; nothing is delivered without an executed agreement and the company's authorization.
- Keep a record of who approved the scope, so a later owner or acquirer can see exactly what was licensed.
Questions to ask your counsel before you sign
- Does the agreement grant a license only, or could any clause be read as assigning rights in our records?
- Which models may the licensee train on our records, and for how long?
- After the term ends, may the licensee keep models trained during it, and must it delete our raw records and any derived material?
- Who owns labels, embeddings, summaries or synthetic data created from our records?
- What must the licensee do to stop outputs reproducing our confidential records?
- May the licensee sublicense, share with affiliates or assign the agreement?
- Do any of our own contracts, notices or policies limit what we can license?
This is general information, not legal, tax or financial advice. Confirm with your own counsel, tax adviser or professional body before acting.
How a referral partner can answer this objection
Owners raise model ownership early, often before they have seen any terms. Keep the reply short and accurate.
A related worry is whether an exclusive license stops the company using its own data; using your data after an exclusive license answers that separately. Sponsors with several portfolio companies sometimes ask about pooling records into one license, which raises its own ownership questions covered in pooling portfolio company data. For background on how buyers source records, see the AI training data marketplace overview.
Next step
If the owner is comfortable with the split, check the basics on the who qualifies page or run the company fit checker. Then register as a partner and make the introduction, or have the owner apply at sourcex.si/apply.
- Step 1Share your linkSend your personal link to a company you know.
- Step 2Company appliesThe company applies itself at /apply.
- Step 3Buyer selects and paysThe buyer selects and pays for the data and SourceX receives its fee.
- Step 4You get your rewardYour share of SourceX fees becomes payable.
Common questions
Can a company ask for a royalty or equity in the model instead of a one-time payment?
Licenses arranged through SourceX use one all-in price paid once, typically within about 60 days of invoicing after the buyer selects the data. A share of model revenue or an ownership stake is not part of that structure. If ongoing participation matters more to an owner than a single payment, raise it before inventory work starts, because it changes which route suits the company.
Can the developer keep using a model after the license ends?
It depends on the post-term and survival clauses. Some agreements let the licensee keep models trained during the term while requiring deletion of the raw records; others limit continued use or require deletion of derived material as well. Ask counsel to read those clauses together, because a duty to delete the dataset says nothing about the model unless the agreement says so.
Could the model repeat our confidential records to other users?
Redaction and de-identification settled before delivery are the main protection, because material that is never delivered cannot be reproduced. The license can add duties on the developer to prevent outputs that reproduce licensed records or identify the company. Ask counsel how those duties are worded and what remedies apply if they are breached, and decide what to leave out of the dataset entirely.
Who owns synthetic data or labels made from our records?
Whoever the agreement names. Owners should not rely on a default rule for derived material such as labels, embeddings, summaries and synthetic records, so the license should define each type, say who owns it, and state whether it must be deleted or may be kept when the term ends. Read the definitions section as closely as the grant clause itself.
Does licensing data for AI training transfer ownership to the buyer?
Not if the agreement is a license rather than an assignment. Under US copyright law, ownership and individual exclusive rights can be transferred separately, so a company can grant a limited training right while keeping ownership of its records. Counsel should confirm that no clause in the agreement could be read as assigning the company's rights.
Related pages
- Can a company still use its own data after signing an exclusive license?
- Can a PE firm combine portfolio company data into a single license?
- AI training data marketplace for licensed company data
- Which US businesses are a fit for a SourceX data licensing introduction
- Check Company Fit for Data Licensing
Free resources
- Days sales outstanding calculator — How many days customers take to pay.
- Business succession planning assessment — Ten questions on successor, transition and documentation.
- NPV calculator — Net present value with a discounted cash flow table.
- All free tools · MCP resource center
By SourceX Partnerships Team · Published 2026-10-09 · Updated 2026-10-09
Know a US company with valuable proprietary data?
Become a referral partner from anywhere we support, get your link and introduce an owner or authorized decision-maker.
Refer a company →I own a business
Explore licensing your company's data to AI developers worldwide. Start a short assessment; no uploads needed.
Start an assessment