The short answer
It depends on the license, not on the buyer's goodwill. The records you delivered may be deleted or kept for the term's survival period, models already trained on them usually keep what they learned, and any onward licenses run on their own terms. The published terms we read say you keep ownership of the underlying data; the agreement decides everything else.
On this page
Who owns the records when the term is over?
Under every published term we read, the selling company never gave ownership away. The license granted rights to use; it did not transfer the records.
| Buyer | Published ownership position |
|---|---|
| micro1 | Underlying data remains the seller’s property |
| Labelbox Prism | Data and rights stay under the seller’s control |
| Nyne | Left to the signed agreement |
| Scale AI | Not spelled out; the page keeps data under the seller’s governance |
Condensed from the four buyers’ own wording, opened 11/07/2026. Disclosure: we can be owed a referral payment by one of these buyers once a company we introduced signs. The table ignores that.
Ownership being yours does not answer the practical questions. Those sit in four places: the delivered copies, the models already trained, the licenses the buyer granted onward, and the money still owed.
What happens to the copies the buyer holds?
Look for a return-or-delete clause. A good one says that within a set period after the term ends, the buyer deletes or returns every copy of the delivered records, including working copies held by its contractors, and certifies that it has done so in writing.
Watch for three common carve-outs:
- Backups. Many agreements let copies sit in backup systems until those roll over. Ask for the rollover period in writing.
- Legal holds. Copies kept because of a dispute or regulation. Reasonable, but they should stay under the same confidentiality terms.
- Derived datasets. Training tasks, labels or summaries built from your records. Whether they count as “your data” for deletion purposes is the single most important definition in the agreement.
What about models already trained on the records?
A model trained on your records keeps what it learned when the data is deleted. There is no practical way to pull one company’s tickets back out of a trained model. Agreements deal with this by limiting what can be trained during the term, not by promising to undo it afterward.
That makes the field-of-use clause decisive. If the license allowed training for any purpose for three years, every model trained in those three years is out of your reach. If it allowed only named uses, the buyer’s options were narrower from the start.
Do onward licenses survive?
Training-data companies exist to supply AI labs, so your records, or tasks built from them, can travel onward. micro1 limits its own customers to labs and enterprises based in the US or allied nations, and promises no resale or distribution while identifying details remain.
Ask two questions of any buyer:
- When the main license ends, do sublicenses end with it, or run to their own term?
- If they run on, does any revenue share continue? Prism’s model pairs a signing fee with a cut of later downstream income; the agreement should fix how long that cut keeps flowing.
Is anything still owed when it ends?
Payment triggers vary. micro1 ties full payment to a 30-day window after acceptance (an FAQ answer counts from the invoice instead). Nyne pays under a license that fixes scope, rights and acceptance tests before any delivery. If a partnership ends early, check whether acceptance had already happened, because that is often the line between paid and unpaid.
For ongoing partnerships, where a company keeps delivering new records, the end of the term also ends future deliveries. Confirm that the final delivery is paid for.
What if the company itself closes?
A closing company’s records do not stop being licensable; they change hands. micro1 keeps a distinct track for defunct firms, open (per its referral rules) only to ones that operated in ten or more countries. In October 2026 MIT Technology Review described labs acquiring records from failed startups; one founder said his code base brought in roughly $25,000.
If your company has an active license and is later acquired or wound down, the license usually passes with the assets. Investors or an acquirer may hold rights to the records after a wind-down. Read the assignment clause.
What should you ask for before signing?
| Clause | What to ask for |
|---|---|
| Return or delete | Deletion of all copies within a set period, certified in writing |
| Derived data | A definition that says whether tasks and labels built from your records are covered |
| Field of use | Named uses only; no “any purpose” grant |
| Sublicenses | Whether they end with the main license |
| Revenue share | How long it continues after the term |
| Survival | Which confidentiality and re-identification promises outlast the term |
Have your own lawyer read the draft. These clauses rarely get attention on the first call and matter most on the last day.
The full deal sequence is on how a deal runs, and the overview of how to sell workflow data shows where the license sits in it.
Read the last page of the license first; the end of a data partnership is decided on the day it is signed.