How to check that an AI data buyer is legitimate
· Company Data Advisors
Six checks an owner can run in an afternoon before sharing records. Who the buyer is, what you pay, what it asks first, personal data, payment and license terms.
Our test: an AI data buyer is legitimate if you can name the company behind it, see how it is funded or regulated, read its process and terms before you share anything, and confirm that you pay nothing to take part. The programs we checked for this guide pass those tests on their published pages. An unsolicited email, a deadline or a promised figure does not tell you anything on its own, so check the buyer, not the message.
This guide gives six checks, what the buyers we track publish on each, and what should make you stop. It is not legal advice; have counsel read any contract before you sign it.
Who is actually behind the offer?
A legitimate buyer is a company you can identify from sources it does not control: funding announcements, regulatory filings, press coverage and a named leadership team. Start there, not with the buyer's own site.
The buyers in our index show what that looks like:
- Public filings. Mode files audited financial statements with the SEC as Mode Mobile, and Appen is listed on the Australian Securities Exchange.
- Reported funding. micro1 committed to a year of buying company data, financed by Citi and Hercules Capital, according to The Next Web. Labelbox Prism comes from Labelbox, which reported a Series E in October 2026 (linked on our buyer page).
- A parent you can look up. Replay is run by the team behind Sunset, which says it has helped thousands of founders wind down companies, and its about page names Sunsets HQ Corp.
- Independent coverage of the category. MIT Technology Review reported on October 13, 2026 that AI labs are buying data from defunct startups.
Stop if the sender cannot tell you the legal name of the contracting company, or if the only evidence it exists is its own website. Also stop if the offer claims a partnership with an institution you can contact and the institution says it has none. Gizmodo reported such a case, in which a university said it was not selling the data an outside group was marketing.
Should a legitimate buyer ever charge me to take part?
We found no fee to apply or sell on any program page we checked, and several say so outright. A fee to apply, to "register" records or to unlock an offer is the clearest sign that something is wrong.
The programs say this on their own pages. Polyshares says the seller pays nothing, and Mode says there is no broker fee. Centific and Appen say there are no setup fees or deductions, and Mercor says there are no setup fees or deductions from the payment and that it covers extraction, anonymization and transfer costs. Our reading of "nothing to pay" covers the buyer's own charges. You may still pay your own lawyer, and a data buyer may ask you to spend staff time on exports and review. Replay says it needs about two hours of your time.
What should a buyer ask for first?
A buyer should ask what you have, not hand you the records. Legitimate programs begin with a description of your systems and a scoping conversation, and they move to a sample only under an agreement.
- Centific and Appen begin with a mutual NDA before any connection. Turing requires a mutual NDA before any data moves, samples included.
- SourceX runs a fit check first that asks for no data uploads.
- The General Data Company signs a confidentiality and data-processing agreement before any sample and moves no file before a signed contract.
- SuperAnnotate holds a scoping call, assesses a sample, values it and signs a written agreement before it prepares and licenses anything.
Stop if a buyer wants logins, bulk exports or customer lists before you have seen a written agreement. Also stop if an intake form asks for personal details of your employees or customers.
Is the estimate a commitment?
An estimate is an early indication, not an offer. Treat any figure as unbinding until it appears in a signed agreement, and treat any buyer that tells you otherwise as one to question.
Buyers publish ranges, and the owner's number depends on the records. Centific gives a written valuation estimate before anything is connected, and Mode sets a fixed offer in a written agreement before export. SourceX records each deal in a Dataset Deal Confirmation covering the dataset, the payout and the deadlines. In these programs the binding number is the one in the agreement. A published range describes what a buyer says it pays across companies, not what yours will get; see what buyers weigh when they value records.
Our view: a deadline is not proof of anything either. A real offer can expire, but you should still have time to read the terms. If a buyer presses you to sign before your counsel has read the agreement, that pressure is itself a reason to slow down.
What happens to personal information in my records?
Your records likely hold employees', customers' and vendors' information. A legitimate buyer states who removes it, when, and where. Compare the published answers before you choose:
| Program | Who removes personal data | Where it happens |
|---|---|---|
| Turing | Turing | On a copy you provide, in an air-gapped environment |
| Centific | Centific | After a read-only connection; you sign off on a sample pack |
| Appen | Appen | Read-only, isolated extraction; you approve a sample |
| Mode | Mode | Your team signs in to approved systems; Mode exports and de-identifies the agreed copy |
| Replay | Replay | An isolated environment for each engagement |
| SourceX | SourceX | Under the agreement, which says no method removes every possibility of re-identification |
Read the SourceX row twice: its agreement states that no method removes every possibility of re-identification, and we cover that in our terms comparison. Whatever the buyer promises, you are still the party who decides whether you may license the records at all. Our view: ask the buyer which categories of personal data it removes, whether it removes them before or after the data leaves your systems, and whether a failure is covered in the contract. If you cannot get written answers, that is your answer. For the law, the Spirit Airlines case and what to leave out, see whether you can license records that include employee and customer data.
When and how am I paid, and what happens to the data afterward?
A legitimate buyer states when payment arrives and what happens to your copy. The programs differ, so compare them against your own priorities.
| Program | Payment as published | Notable term |
|---|---|---|
| Turing | A share of each sale, nothing upfront | Non-exclusive; Turing sells alone for the first 90 days, then you can renew or opt out |
| micro1 | Within 30 days of acceptance, per its program page | Raw data deleted on an agreed schedule or on request |
| Polyshares | Net 30 to 60 once data is approved, shared and anonymized | Published deals are perpetual exclusive licenses |
| SourceX | Within 60 days of receiving your invoice, issued after closing | Exclusive for AI training, for a term set in each deal |
| Replay | Upfront payment plus a revenue share | Perpetual revenue share |
| Labelbox Prism | Upfront fee at signing plus a revenue share | Exclusive by default |
All of this comes from the buyers' own pages as we last checked them. The payment timing tells you whether a buyer is organized; the license terms tell you what you are giving up. Several of these deals are perpetual or exclusive, and deleting raw data does not undo a model already trained on it. Our terms comparison covers exclusivity, term, resale and what survives in detail. Stop if the buyer will not put the payment date, the amount and the license term in the same written agreement.
What a legitimate program lets you do
You should be able to read the process before you apply, apply to more than one buyer, and decline after hearing the terms. How to sell or license your company's data walks through the steps, and how the market for company data works explains why a lab rarely buys from you directly.
Where we come in
Company Data Advisors is an independent advisory firm, not a broker. We help owners compare programs and see where their records would go, and nothing is shared until the owner approves a specific, named buyer. Participating buyers may pay us referral fees. Owners can register with any buyer directly, and the six checks above work the same way whether you use us or not.
What to do this week
- Write down the legal name of each buyer you are considering, and find one source about it that the buyer did not write.
- Confirm in writing that you pay nothing to apply or to sell.
- Ask for the process and the agreement before you share a login or a file.
- Ask who removes personal data, and when.
- Ask for the payment date, the license term and exclusivity in one document, and have your counsel read it.
If you want a second set of eyes on a specific offer, tell us what you have and we will walk through the same checks with you.
Programs in this post
More from the blog
- Is my company too small to license its data to AI buyers?
Most AI data buyers with a size floor set it at 20 to 50 employees, but law, accounting and medical firms qualify smaller, and nine programs state no minimum.
- What company data AI buyers want: November 2026
Across the 17 AI data-buyer programs we track, accounting, healthcare and software lead the industries, and CRM, ticket, email and finance records lead.
- One payment or a revenue share: how AI data buyers pay
The largest group of AI data buyers pays once. A few pay a share of every resale, and one now pays nothing upfront. How each model works and what to ask.
See which buyers fit your company
Eight quick questions against every program in our buyer index. Participating buyers may pay us referral fees.