# Vaultfish: Guides and the price study

> 22 guides and the price study, as markdown. One part of https://vaultfish.com/llms-full.txt. Index of everything published: https://vaultfish.com/llms.txt

Cite the canonical URL each document names, never this file. Nothing here is legal advice. The fit check, a discovery call, tools, guides and introductions are free to your company. Managed readiness is optional, scoped and quoted in writing before you pay. micro1 may pay us a referral fee if a company we introduce goes ahead with it. We never take a percentage of your deal, and we do not negotiate for you. https://vaultfish.com/disclosure explains how we are paid.

---

# How much is your company data worth to AI?

> No price list exists for company data. What moves its value, what public deals and buyer calculators show, and a ten-minute check to run on your own records.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-much-is-your-business-data-worth-to-ai  
Published: 2026-10-07  
Updated: 2026-10-14

Nobody can give you a reliable price for company data today. The public record holds a handful of deals and a few buyer calculators that disagree widely. It also holds a short list of things that move the number.

> **Data licensing means a company gets to use your records for a while, under rules. You keep ownership.** Think of the attic. You can't price the boxes until you know what is in them, how old they are and whose they are.

A ten-minute look at your own records will tell you more than any estimate form. The check is at the bottom. The full process is in the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies).

## Why is there no rate card for what company data is worth?

[Troveo](/buyers/troveo) says there is [no public price list](https://www.troveo.ai/resources/sell-data-to-ai-companies), and no public per-record, per-token or per-workflow-hour rate for business data exists yet.

[micro1](/buyers/micro1) states deal tiers. It also says [compensation depends](https://www.micro1.ai/data-partnerships) on dataset size, workflow complexity, data quality and domain expertise.

Part of the reason is that the buyer side shares little detail. Marketplaces name buyers in generic terms. Epoch AI's review of RL environments describes app replicas and lab partnerships with product companies, and does not discuss real company archives ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)).

So every number below is a deal, a bid or a company's own claim. None is a rate you can multiply by your headcount.

## What moves the value?

**History.** Buyer calculators take headcount and years of history as inputs. Count the period your records actually cover, not the company's age.

**Volume.** The Spirit archive was described in counts: about 100 million emails, 500 million Teams chats and about 30 million lines of code ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)). Counts are the language of the conversation.

**Context.** This one is our read, and no published price proves it yet. micro1's CFO and Troveo's guide both point to records of real work: a problem, the steps and the result. Whole threads that show that arc should be worth more than loose fragments.

**Rights.** A record you have no right to license has no value to a buyer, so rights set the ceiling. Counsel's list covers third-party confidentiality, trade secrets, privilege and employee talk the US labor law protects ([counsel's list](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

**Buyer demand.** Bids on Spirit rose from $7.5M to $12.5M. That points to demand. Supply is rising too. More than 350 companies have listed assets on SimpleClosure's hub since April ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)).

**Exclusivity.** Epoch's interviews put exclusive deals at roughly 4x to 5x the non-exclusive price for RL environments. Troveo says its exclusivity is time-limited and has to be earned.

## What do the public deals show?

| Evidence | Figure | Read it as |
|---|---|---|
| Google, Spirit Airlines archive, August 2026 | $10M winning bid, still contested | One airline's archive, sale awaiting court approval |
| micro1, Spirit, September 2026 | $12.5M later offer | Same archive, a related offer and not a separate deal |
| SimpleClosure, 2026 | $10,000 to $100,000 per shut-down startup, nearly 100 deals | Small startups ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)) |
| Cielo24, a shut-down captioning company's Slack messages, email and Jira tickets | "Hundreds of thousands of dollars" | One reported case, from Forbes via [Slashdot](https://yro.slashdot.org/story/26/04/18/014244/shuttered-startups-are-selling-old-slack-chats-emails-to-ai-companies) |
| Mercor, shuttered startups | Offers reported up to $300,000 per startup | [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d) |
| micro1 | Tiers of $100K+, $500K+ and $1M+, and $100K-$2M+ for approved data packages | Self-reported, from [its page](https://www.micro1.ai/data-partnerships) and its [Enterprise Data Partnership listing](https://www.micro1.ai/company-referral) |
| Troveo | Full-company deals "start at six figures" | Self-reported ([Troveo](https://www.troveo.ai/business-data)) |
| Epoch's RL environment review | About $20,000 for a website clone, figures around $300,000 for a Slack-grade clone, $200 to $2,000 per task | What labs pay to build environments, a different product from your archive |

**$10,000 to $100,000** What shut-down startups got per company, across nearly 100 deals ([source](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916))

The Spirit sale is still before the bankruptcy court, with no sale order. The flight attendants' and pilots' unions objected. EPIC filed a brief on 2 Oct 2026 ([EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)).

On 5 Oct 2026 the court-appointed privacy ombudsman recommended approval from a privacy view. Her report says she did not examine risks to employees or third parties ([ombudsman report](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf), [Bloomberg Law](https://news.bloomberglaw.com/ip-law/spirit-data-sale-to-google-passes-muster-privacy-ombudsman-says)).

The hearing was moved from 14 Oct to 6 Nov 2026 ([Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), 9 Oct 2026). No sale order as of 10 Oct 2026.

The [deals tracker](/deals) follows each public figure.

## Why do calculators disagree on what business data is worth?

Each calculator runs on its own assumptions, so the same company gets very different answers. Troveo's calculator shows $800K to $1M by default, with employee tiers from 20 to 1,000+ ([Troveo](https://www.troveo.ai/business-data)).

Other estimate forms I tested gave ranges far from that for similar inputs, and some showed a number before I entered anything.

These tools run on headcount, founding year and sometimes revenue. micro1 says pay depends on dataset size, workflow complexity, data quality and domain expertise ([micro1](https://www.micro1.ai/data-partnerships)). A headcount form asks for none of those.

Most calculators sit in front of a lead form that collects your name, email, phone and website, then books a call. That is how the buyer finds its next deal. Use a calculator to start a conversation. Treat the number as a hint about the buyer's pitch.

_A calculator that gives you a number before it asks what you have is guessing. Politely._

## Run the ten-minute readiness check

Set a timer and open a blank doc. Write one block per system, using the template below.

1. **Systems.** List every tool where work gets recorded: email, chat, CRM, helpdesk, project boards, docs, call recordings, code. Note who can export each one.
2. **Years.** Write the date of the oldest record you still hold in each system, and any gaps.
3. **Counts and size.** Estimate the number of items per system, such as tickets, threads, deals and messages. Add the storage each system uses.

   micro1's pages state no data-size threshold, but a storage count helps you describe your records. Screenshot wherever the tool shows a total, with the date visible. Then you can show evidence, not a recollection.
4. **Rights.** Write whose words are in each system: staff, customers, partners, contractors. Note which contracts mention confidentiality or data use. Note whether your tool's terms restrict export or AI use.
5. **Exclusions.** Mark what stays out: anything under NDA, trade secrets, privileged conversations and employee talk the US labor law protects. Mark personal data about customers too. Removing names from it is harder than it sounds.

```text
System: [tool name]
Oldest record: [date]
Approximate count: [number and unit]
Whose words: [staff / customers / partners / contractors]
Restrictions: [contracts, tool terms, privacy policy]
Exclude: [what stays out]
Screenshot saved: [yes / no]
```

Don't change your privacy policy or customer terms to cover this after the fact. The FTC calls quiet retroactive changes for AI training potentially unfair or deceptive ([FTC](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive)). Ask a lawyer how to handle your own terms.

## What to do with the sheet

Bring it to a lawyer first, because rights decide what is on the table at all. Then bring it to a buyer or data program. The stated thresholds differ.

micro1's pages say 30 or more employees, with mature operations. Handshake AI's referral page says 20 or more full-time employees, and Troveo states no minimum. Check yours.

The free [fit check](/business) is the next step. The [buyer directory](/buyers) lists the buyers side by side.

If your business runs a community, a course library or a podcast, the people in that content hold rights too. The [guide to licensing your business's content to AI](/guides/license-your-business-content-to-ai) covers members, clients and guests.

The pillar guide has the [contract checklist and red flags](/guides/how-to-sell-company-data-to-ai-companies) to read before any call. If a buyer quotes you a number, ask what it is built from.

## Do this safely

Run the [rights check](/rights-check) before any buyer sees your sheet. It asks seven questions about whose words sit in your systems and what your contracts say.

The [risk map](/risks) covers the legal and security failure points for data owners. Keep raw exports in your own systems until a signed agreement says who receives them.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [no public price list](https://www.troveo.ai/resources/sell-data-to-ai-companies)
- [compensation depends](https://www.micro1.ai/data-partnerships)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [counsel's list](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)
- [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [Slashdot](https://yro.slashdot.org/story/26/04/18/014244/shuttered-startups-are-selling-old-slack-chats-emails-to-ai-companies)
- [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)
- [Enterprise Data Partnership listing](https://www.micro1.ai/company-referral)
- [Troveo](https://www.troveo.ai/business-data)
- [EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)
- [ombudsman report](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)
- [Bloomberg Law](https://news.bloomberglaw.com/ip-law/spirit-data-sale-to-google-passes-muster-privacy-ombudsman-says)
- [Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data)
- [FTC](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive)

---

# How to license your company's data to AI companies

> Searching how to sell your company's data to AI? You license it. Shut-down startups got $10,000 to $100,000. Who pays, what qualifies and the rights checks.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-to-sell-company-data-to-ai-companies  
Published: 2026-10-07  
Updated: 2026-10-14

Yes, some AI companies pay for company records, but the public proof is thin. The one repeat price is $10,000 to $100,000 per shut-down startup, per [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916).

People search for how to sell company data. In practice you license it. A working company has no public price yet. Most deals run through a marketplace, after you confirm you hold the rights.

**$10,000 to $100,000** What SimpleClosure's nearly 100 deals paid per shut-down startup ([source](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916))

That range is for a company closing its doors. No raw data leaves your building until a contract says so.

The buyer numbers are self-reported, and this guide leans on public reporting, court filings and buyer pages. Check anything you plan to rely on.

## What do AI companies want from a business?

Troveo, micro1 and Mercor list records of real work. A support ticket shows a problem, the steps taken and the result. A deal in your CRM (the system that tracks customers) shows the same arc.

Labs and agent builders use records like these to build practice workplaces for AI agents, sometimes called RL gyms. The agent works through real Slack threads, tickets and email ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916), [Futurism](https://futurism.com/artificial-intelligence/employer-selling-emails-train-ai)).

Most marketplace pages keep buyers generic, as in "frontier AI lab." For the limits of this market, see [what can go wrong](/what-can-go-wrong).

## What qualifies, and what stays out?

The companies list the same core material:

- [micro1](https://www.micro1.ai/data-partnerships): written procedures, knowledge bases, CRM data and project histories.
- [Troveo](https://www.troveo.ai/business-data): chat, email, documents, tickets, CRM and finance systems.

Size and history matter too. micro1 asks for 30 or more employees, Handshake AI for 20 or more full-time employees and 3 or more years in operation, and Troveo states no minimum. The table shows where each says so:

| Company | Stated minimum | Where it says so |
|---|---|---|
| micro1 | 30 or more employees, mature operations, primarily English documentation, US prioritized | [Data partnerships page](https://www.micro1.ai/data-partnerships) and its company referral page |
| Handshake AI | 20 or more full-time employees and 3 or more years in operation | [Referral page](https://joinhandshake.com/ai/data-partnerships/referral) |
| Troveo | No minimum stated | [Troveo](https://www.troveo.ai/business-data) |

Acceptance at micro1 depends on its review and current demand. Confirm exceptions directly with micro1.

What stays out is anything you lack the right to license. Your email and chat hold other companies' information, often under NDAs (confidentiality agreements).

Counsel also flags trade secrets, waived privilege and employee talk protected by US labor law ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

Your tools have terms too. Salesforce's 2025 Slack API terms bar bulk export and AI use of data pulled through its API (the connection other tools use) ([Hunton](https://www.hunton.com/insights/legal/salesforce-locks-down-slack-data-time-to-review-your-slack-api-terms)). Ask whether an admin's own export is treated differently.

## How does a deal work?

Steps vary by buyer. This is the common shape.

1. **Assessment** List your tools, years of records and rough counts.
2. **Scope** Agree in writing which systems and dates are in or out.
3. **Preparation** Someone extracts the records and removes names.
4. **Contract and transfer** Sign first. Then the data moves.
5. **Payment**

Assessment usually starts from counts and dates, not raw records. [Troveo](/buyers/troveo) says it needs no OAuth into your systems and no platform to install ([Troveo](https://www.troveo.ai/business-data)). Many buyers also run a short estimate form, then a call.

Preparation differs a lot. Troveo says it scrubs names, contact details and client identities, and needs no sign-in access to your systems. Others connect read-only to the tools you approve.

[micro1](/buyers/micro1)'s CFO said in an undated recorded forum talk that its engineers help your IT people with the transfer, about 10 hours of work over a week or two from one or two of your IT people.

micro1's David Revland, who works on human data, said micro1 verifies the data before it pays.

Revland also said micro1 swaps each name for a consistent stand-in, so roles and cross-tool trails survive. Its page says personal information "is removed when relevant." Ask each buyer which approach it uses and what it means for your people.

Payment comes in four models:

- **One-off transfer.** [SimpleClosure](/buyers/simpleclosure), whose terms give the buyer all right and title.
- **Direct partnership.** micro1's corpus track covers a full archive of company records, paid under the agreed partnership. micro1 says companies keep ownership of their underlying data.
- **Revenue share.** Troveo pays a cut of repeat deals. The split is unpublished.
- **Deal revenue share.** Protege pays a revenue share when your content enters qualified deals ([Protege](https://withprotege.ai/data-provider/video)).

Exclusivity changes the price. Epoch AI's interviews put exclusive deals at roughly 4x to 5x the non-exclusive price. [Troveo](https://www.troveo.ai/business-data) says its exclusivity is time-limited.

## What do the public figures show?

There is no public price list, and [Troveo's own guide says so](https://www.troveo.ai/resources/sell-data-to-ai-companies). What exists is a short list of deals.

- **Spirit Airlines archive.** Google's $10M bid won the 14 Aug 2026 auction, and court approval is pending.

  The archive held about 100 million emails, 500 million Teams chats and 30 million lines of code. [Mercor](/buyers/mercor) was the alternate at $7.5M ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)). micro1 later offered $12.5M ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)).
- **SimpleClosure.** Nearly 100 deals in a year at $10,000 to $100,000 per shut-down startup ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)).
- **Mercor.** Offers to shuttered startups reported at up to $300,000 each ([BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)).
- **micro1 and Troveo.** micro1 lists $100K+, $500K+ and $1M+ tiers. Troveo says full-company deals typically start at six figures.

**The biggest number is still a bid:**

The court has not approved the Spirit sale. Unions objected, saying stand-in names could let someone rebuild identities. EPIC filed a brief on 2 Oct 2026 ([EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)).

The court-appointed privacy ombudsman said on 5 Oct 2026 that approval would be appropriate from a privacy view. Her report says she did not reach any conclusion on risks to employees or third parties ([ombudsman report, Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf), [Bloomberg Law](https://news.bloomberglaw.com/ip-law/spirit-data-sale-to-google-passes-muster-privacy-ombudsman-says)).

The debtors' notice ([Doc 1663](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1663.0.pdf)) first set the hearing for 14 Oct 2026, and it was later moved to 6 Nov 2026 ([Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), 9 Oct 2026). No sale order as of 10 Oct 2026.

Every public deal above is a distressed company selling a whole archive. None shows what a working company would get. The [value guide](/guides/how-much-is-your-business-data-worth-to-ai) covers what moves the number, and the [deals tracker](/deals) follows the public figures.

## Which companies take company data?

Troveo is a marketplace that pays a revenue share. micro1 and Mercor deal with companies directly. SimpleClosure serves shutdowns.

The [buyer directory](/buyers) covers each player neutrally, for all of them and more.

Some companies in this market also pay people who introduce a business.

Ask any comparison, this one included, whether its author gets paid when you sign up. Ours may be, and the [disclosure page](/disclosure) says how.

## Do you have the right to license your company's records?

The largest risk sits in the archive. Your customers and employees wrote those messages, and your contracts may not let you license them.

Removing names is harder than it sounds. The UK data regulator says pseudonymised data is still personal data for anyone who holds the extra information ([ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)). A label like "Person 14" keeps the links between records.

Law differs by place. The EU's [EDPB Opinion 28/2024](https://www.edpb.europa.eu/system/files/2024-12/edpb_opinion_202428_ai-models_en.pdf) requires anonymity claims to survive a test of "all the means reasonably likely to be used." The FTC calls quiet retroactive changes to terms for AI training potentially unfair or deceptive ([FTC](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive)).

Whether you need consent from employees and customers depends on your contracts and your state. That question comes before any buyer call.

## How do you do this safely?

Three steps before any call, in this order:

1. **Run the [rights check](/rights-check).** Seven questions on whose words sit in your systems and what your contracts say.
2. **Read the [risk map](/risks).** It maps the legal and security failure points for data owners.
3. **Hold your raw data.** Keep exports in your own systems until a signed agreement says who receives them and when they are deleted.

Check each buyer's security record too. Mercor says a supply chain attack in March 2026 exposed sensitive information for a very limited subset of its experts ([Mercor](https://www.mercor.com/blog/update-on-mercor-security-incident/)).

## What contract questions go to a lawyer?

We are not lawyers, so read this as a list to bring to one.

1. **Scope.** Which systems and dates, and what is excluded.
2. **Payment.** One-off, revenue share or royalty, with the trigger in writing.
3. **Exclusivity.** How long, for which buyers, at what premium.
4. **Removing names.** Who does it, to what standard, and whether links survive.
5. **Raw data.** When it moves, over what access, and when it is deleted.
6. **Liability.** Who pays if a record leaks or a rights claim lands.
7. **Rights promises.** What you warrant about ownership and consent.
8. **Buyer limits.** What the buyer may do with the data, and whether it can resell.
9. **Term and exit.** How long the license runs and how it ends.
10. **Referral fees.** Whether anyone between you and the buyer takes a cut.

## What are the red flags?

- **A payout clock.** A promise to pay within days usually starts once your data passes review, not at the first call. Ask what starts the clock.
- **Headline ranges.** A headline range can sit far above what the same site's calculator returns. Ask what the range is built from.
- **Urgency.** Posts saying owners who wait will get less cite no source. More than 350 companies have listed on SimpleClosure's hub since April, so supply is rising too ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)).
- **No terms or privacy policy.** If a buyer's terms or privacy page returns "not found," ask for the contract. Do not give any buyer read access before you have read it.

## Where do you start this week?

Run the ten-minute check in [how much your business data is worth to AI](/guides/how-much-is-your-business-data-worth-to-ai). It gives you the counts every buyer will ask for. Then the free [fit check](/business) is the next step.

If your business runs a community, a course library or a podcast, read [the guide to licensing your business's content to AI](/guides/license-your-business-content-to-ai) first. Members and guests hold rights the business may not.

Then put the contract questions above in front of a lawyer before any buyer call.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [Futurism](https://futurism.com/artificial-intelligence/employer-selling-emails-train-ai)
- [micro1](https://www.micro1.ai/data-partnerships)
- [Troveo](https://www.troveo.ai/business-data)
- [Referral page](https://joinhandshake.com/ai/data-partnerships/referral)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Hunton](https://www.hunton.com/insights/legal/salesforce-locks-down-slack-data-time-to-review-your-slack-api-terms)
- [Protege](https://withprotege.ai/data-provider/video)
- [Troveo's own guide says so](https://www.troveo.ai/resources/sell-data-to-ai-companies)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)
- [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)
- [EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)
- [ombudsman report, Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)
- [Bloomberg Law](https://news.bloomberglaw.com/ip-law/spirit-data-sale-to-google-passes-muster-privacy-ombudsman-says)
- [Doc 1663](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1663.0.pdf)
- [Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data)
- [ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)
- [EDPB Opinion 28/2024](https://www.edpb.europa.eu/system/files/2024-12/edpb_opinion_202428_ai-models_en.pdf)
- [FTC](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive)
- [Mercor](https://www.mercor.com/blog/update-on-mercor-security-incident/)

---

# AI training data: where labs get it

> AI training data is what models learn from and are tuned and tested on. Where labs get it, why public text is limited, and what company records add.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/ai-training-data  
Published: 2026-10-01  
Updated: 2026-10-14

AI training data is the text, images, audio, video, code and records that a model learns from. Labs also use it to tune and test the model.

Labs get it from six places: scraped web data, open datasets, platform and user data, made-up data, direct licensing deals and licensed data providers.

Company records are the newest source. They add something public text lacks, a record of real work from start to finish.

> **AI labs have read most of what is public.** Think of a library. The shelves are nearly read. Now the labs want to borrow how real work gets done.

This guide covers the sources, the limits on public text and what a company's own archive adds. The licensing process lives in the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies).

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

## What counts as AI training data?

Three jobs use it. Pre-training teaches a model from very large amounts of data. Post-training and fine-tuning shape its behavior with smaller, curated sets. Evaluations test it.

Protege's pages describe its datasets as for "pre-training, post-training, fine-tuning and evals."

Human work sits inside the second and third jobs. Experts write model answers, grade output and flag errors. Platforms such as Mercor, Surge AI, Scale AI and Turing pay people to do it. That is labor sold to a lab. Your existing records are a different product.

## Where do labs get their data?

[Troveo](/buyers/troveo)'s guide lists six sources: scraped web data, open datasets, platform and user data, synthetic (made-up) data, direct licensing deals and licensed data providers ([Troveo](https://www.troveo.ai/resources/where-ai-labs-source-training-data)). Troveo sells licensed data, so read its ranking with that in mind.

The same guide says made-up data fills gaps but "inherits the ceiling of the models that generate it." Labs use it as a supplement.

On platforms, labs with their own user content use it. Labs without it pay for someone else's. Reuters reported in February 2024 that Google pays Reddit about $60M a year ([The Decoder](https://the-decoder.com/reddit-signs-60-million-annual-training-data-deal-with-google/)).

Shutterstock's CEO said its AI licensing made $104M in 2023 ([Bloomberg Law](https://news.bloomberglaw.com/artificial-intelligence/shutterstocks-ai-licensing-business-generated-104-million)).

## Is public training text running out?

Epoch AI estimated that the supply of good public human text could be fully used between 2026 and 2032, an 80% interval ([Epoch](https://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-data)).

It put the supply near 300 trillion tokens. It said heavier training on the same data could pull the date earlier.

The range is wide, so treat the dates as a bracket. The count covers public text only. Licensed libraries and private company records sit outside it. That is why private data gets attention.

## What does company data add?

It adds a record of how a real problem got solved. Who asked, which tools were used, what approvals came and what the result was.

micro1's CFO described the prize in an undated recorded forum talk as the chain from a communication to a deliverable.

David Revland, who works on human data at micro1, gave the example of a Slack message asking for a finance report, then the report in a spreadsheet.

Fortune calls the recorded steps a trajectory ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)).

Labs and vendors use such records to build RL environments. These are simulated workplaces where an agent practices a job ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)).

Google bid $10M for the Spirit Airlines archive. It held about 100 million emails, 500 million Teams chats and about 30 million lines of code ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)). The sale was still before the bankruptcy court.

## What does the demand look like?

Epoch interviewed RL environment vendors. A website-clone environment cost near $20,000. A Slack-grade clone cost near $300,000. Tasks ran $200 to $2,000. Exclusive deals ran roughly 4x to 5x the non-exclusive price ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)).

Those prices pay for environments and tasks that someone built. That differs from paying for an existing archive. Epoch's review does not discuss real company archives.

That is the gap in the demand case. Marketplace copy rarely names a buyer.

**Figure: A practice Slack costs about fifteen practice websites to build.** Two rows of squares for the price of one RL environment in Epoch AI's vendor interviews, each square worth $20,000. A practice website clone near $20,000 is one square. A practice Slack-grade clone near $300,000 is fifteen highlighted squares. Source: [Epoch AI vendor interviews](https://epoch.ai/gradient-updates/state-of-rl-envs).

## What happens when rights are unclear?

Content has already produced large claims. Anthropic agreed in September 2025 to pay at least $1.5B to settle an authors' lawsuit.

The suit covered about 500,000 books that the plaintiffs said came from piracy sites. That works out to roughly $3,000 per book ([NBC News](https://www.nbcnews.com/tech/tech-news/anthropic-settlement-lawsuit-authors-rcna229422)).

That case is about books, and company records are a different kind of material. It shows why buyers and data owners both care where data came from and who agreed to its use.

For company archives, the live dispute is the Spirit sale. The flight attendants' union and EPIC objected on privacy grounds ([EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)).

## What is the owner's part?

Rights come first. Your archive holds customers' and employees' words. Counsel flags third-party confidentiality, trade secrets, legal privilege and employee talk that US labor law protects ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

Run the [rights check](/rights-check) and read the [risk map](/risks).

Count your systems and years of records using the sheet in [how much your business data is worth to AI](/guides/how-much-is-your-business-data-worth-to-ai). Then take the free [fit check](/business). The [AI data licensing guide](/guides/ai-data-licensing) explains outright transfer versus license. [AI data companies](/answers/ai-data-companies) sorts the players.

_Your old Slack has years of real work in it. The library has never seen it. Check the rights first._

Talk to a privacy or commercial lawyer in your state before you sign.

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link: micro1 may pay us a referral fee if a company we introduce goes ahead with it. You pay nothing extra._

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Troveo](https://www.troveo.ai/resources/where-ai-labs-source-training-data)
- [The Decoder](https://the-decoder.com/reddit-signs-60-million-annual-training-data-deal-with-google/)
- [Bloomberg Law](https://news.bloomberglaw.com/artificial-intelligence/shutterstocks-ai-licensing-business-generated-104-million)
- [Epoch](https://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-data)
- [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)
- [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [NBC News](https://www.nbcnews.com/tech/tech-news/anthropic-settlement-lawsuit-authors-rcna229422)
- [EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)

---

# How AI labs use company data

> How AI labs use company data: what arrives, how names get swapped, what labs build with it, what you keep and what you can't take back. Buyer by buyer.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-ai-labs-use-company-data  
Published: 2026-09-30  
Updated: 2026-10-14

AI labs use company data to train and test software agents. Your Slack threads, email chains, tickets and documents show how a real task moved from request to finished deliverable. Public text can't show a lab that path.

The data usually arrives as an export or a connected account. A buyer swaps the names. A lab then builds practice tasks and scoring tests from it.

> **Data licensing means a lab gets to use your records for a while, under rules.** Think of a library. The lab borrows real work to learn how work gets done. You keep the originals.

You can't get a trained model back out of a lab. Most questions below trace to that.

The public record on what happens inside a lab is thin. Buyers describe their own pipelines on their own pages, so those claims are self-reported. Where a buyer says nothing, we say so.

## What data actually arrives?

Three collection methods show up across the buyers we read.

[Mercor](/buyers/mercor) connects to your accounts. [Its page](https://www.mercor.com/data/) lists 100+ supported tools and says no data leaves your systems until you approve it.

[Troveo](https://www.troveo.ai/business-data) asks for an export and says it needs no OAuth (a login link) into your systems. [micro1](/buyers/micro1) starts with a form that lists your tools and rough sizes. Its engineers then help your IT people with the transfer.

The content is workplace records. [micro1 lists](https://www.micro1.ai/data-partnerships) SOPs, knowledge bases, CRM data, project histories and QA processes. [Troveo](/buyers/troveo) lists chat, email, documents, work tools, CRM and support, finance, and code on its business data page.

micro1's CFO says labs want the chain: a request in chat, the spreadsheet that answered it, and the approval that followed. The CFO used a marketing budget request as the example in an undated recorded forum talk.

Each buyer has its own guide, linked at the bottom. Start with [how micro1 handles it](/guides/how-micro1-uses-your-company-data) or [how Mercor handles it](/guides/how-mercor-uses-your-company-data).

## How do the companies change the data before a lab sees it?

Every buyer says it removes or swaps personal details. The methods differ, and so does what is left behind.

Troveo shows the cleanest example on its [business data page](https://www.troveo.ai/business-data). A sentence naming a customer and an email address becomes a sentence with PERSON-014 and EMAIL-019 in their place.

micro1's David Revland, who works on human data, describes a similar swap in an undated recorded forum talk. A name becomes the same stand-in on every platform, so the role and the cross-tool trail survive.

[micro1's page](https://www.micro1.ai/data-partnerships) says some datasets get synthetic rewrites. micro1 staff said in the same talk that companies can't clean the data first, because blanking names destroys the trail it needs.

[Mercor](https://www.mercor.com/data/) says its extraction and anonymization is automated. [Handshake AI](https://joinhandshake.com/ai/data-partnerships/) says it applies a structured process to remove direct and indirect identifiers.

A swapped name can still point to a person. The [UK Information Commissioner's Office](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/) says pseudonymised data is personal data in the hands of anyone who holds the extra information. [Handshake AI](/buyers/handshake-ai) says on its own page that no process of removing names eliminates re-identification risk entirely. The detail is in [micro1's name transformation](/answers/micro1-pii-transformation) and [pseudonymization versus anonymization](/answers/pseudonymization-vs-anonymization).

Figure: steps Export or connect -> Lab trains or tests -> Model keeps what it learned. Settle what the model may keep before you export.

## What do labs build with it?

Public reporting names three uses: reinforcement learning environments, agent training and evaluations.

An [RL environment](/answers/what-is-an-rl-environment) is a practice space. A model attempts a task and an automated grader scores it. Time tied Google's Spirit Airlines bid to environments like that. Google told Time only that the data "can be helpful in improving our products and AI models."

Fortune, reporting on micro1's offer for Spirit's records, wrote that an agent needs to learn how to choose actions, use tools and recover from mistakes. The record of those steps is a [trajectory](/answers/what-is-a-trajectory-in-ai).

The third use is evaluation. micro1's workflows track has employees grade a general-purpose agent on their real daily work. micro1 receives the grade plus the agent's trajectory.

The gap: no lab has published how it used a specific purchase. We have buyers describing plans. The longer read is [what AI labs do with company data](/answers/what-do-ai-labs-do-with-company-data).

## What do you keep?

The companies split on this in their own words.

Figure: comparison of What each company says about your data rights: micro1 (You retain ownership), Troveo (Ownership never transfers), SimpleClosure (emphasized: Perpetual and irrevocable, Perpetual license). Two say you keep ownership. A perpetual license never ends, so check the contract.

[micro1's page](https://www.micro1.ai/data-partnerships) says companies retain ownership of their underlying data. [Troveo's licensing guide](https://www.troveo.ai/resources/exclusive-vs-non-exclusive-data-license) says a license is permission and you keep ownership. [SimpleClosure's terms](https://simpleclosure.com/tos/) take a non-exclusive, transferable, sub-licensable, perpetual and irrevocable license to the assets you submit.

Owning your files and the rights a buyer holds over its copy are separate questions. Read the contract for the second one. Our page on [what reps and warranties a buyer asks for](/answers/what-reps-and-warranties-does-an-ai-data-buyer-ask-for) shows what you will be asked to promise.

## What can't you take back?

Deleted files don't un-train a model. [Troveo's guide to licensing deals](https://www.troveo.ai/resources/ai-data-licensing-deals) says models already trained survive the license ending. A [law firm checklist](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes) from Frankfurt Kurnit Klein and Selz says data that has trained a model cannot practically be pulled back out.

A regulator can reach further than a contract can. In February 2024 the [FTC's order against Avast](https://www.ftc.gov/news-events/news/press-releases/2024/02/ftc-order-will-ban-avast-selling-browsing-data-advertising-purposes-require-it-pay-165-million-over) required deletion of the browsing data sent to Jumpshot and of the algorithms derived from it. That was a consumer privacy case and not a company-archive deal. It still shows that a model is not always safe from an order.

What a buyer will delete is a contract term. micro1 says data is retained only for a limited period under agreed retention policies, with deletion on request. Mercor's page says nothing on retention or deletion. [Deleting after training](/answers/can-i-delete-my-data-after-an-ai-company-trains-on-it) covers the three separate questions inside this one.

_Deleting a file is easy. Un-teaching a model is a different job, and nobody has the button._

## Where can it leak?

Four places, in the order we would worry about them.

1. **The buyer's own security.** [Mercor said](https://www.mercor.com/blog/update-on-mercor-security-incident/) a malicious release of the open-source LiteLLM library in March affected it, and that the impact on customer information was very limited. [The Next Web](https://thenextweb.com/news/meta-mercor-breach-ai-training-secrets-risk) reported about 4 TB exposed, including source code, a user database and interview recordings. I found no report of company-archive data exposed. The facts are on [the Mercor data breach page](/answers/mercor-data-breach).
2. **Re-identification.** Swapped names plus context can point back to a person.
3. **Rights you did not hold.** Your customers' and employees' messages are inside the archive. You promise the buyer you had the right to share them.
4. **The people who can see files before the swap.** Ask who that is, in writing.

Who pays if a buyer is hacked is usually a contract term. See [who is liable if the data buyer is hacked](/answers/who-is-liable-if-the-data-buyer-is-hacked).

## Which companies explain their handling?

Six publish enough to read. Each guide covers what the company asks for, how it collects, what it says it does to the data and who gets it. It also covers ownership, retention, security, how it pays and what to ask.

- [How micro1 uses your company data](/guides/how-micro1-uses-your-company-data)
- [How Mercor uses your company data](/guides/how-mercor-uses-your-company-data)
- [How Troveo uses your company data](/guides/how-troveo-uses-your-company-data)
- [How Handshake AI uses your company data](/guides/how-handshake-ai-uses-your-company-data)
- [How SimpleClosure uses your company data](/guides/how-simpleclosure-uses-your-company-data)
- [How Defined.ai uses your company data](/guides/how-defined-ai-uses-your-company-data)

The [buyers directory](/buyers) lines them up side by side. Before any call, run the [rights check](/rights-check) and read the [risk map](/risks).

Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Its page](https://www.mercor.com/data/)
- [Troveo](https://www.troveo.ai/business-data)
- [micro1 lists](https://www.micro1.ai/data-partnerships)
- [Handshake AI](https://joinhandshake.com/ai/data-partnerships/)
- [UK Information Commissioner's Office](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)
- [Troveo's licensing guide](https://www.troveo.ai/resources/exclusive-vs-non-exclusive-data-license)
- [SimpleClosure's terms](https://simpleclosure.com/tos/)
- [Troveo's guide to licensing deals](https://www.troveo.ai/resources/ai-data-licensing-deals)
- [law firm checklist](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [FTC's order against Avast](https://www.ftc.gov/news-events/news/press-releases/2024/02/ftc-order-will-ban-avast-selling-browsing-data-advertising-purposes-require-it-pay-165-million-over)
- [Mercor said](https://www.mercor.com/blog/update-on-mercor-security-incident/)
- [The Next Web](https://thenextweb.com/news/meta-mercor-breach-ai-training-secrets-risk)

---

# AI data licensing, and who actually pays for it

> AI data licensing lets AI companies use your data on set terms. Outright transfer vs license, four kinds of data, who pays, and what an owner needs first.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/ai-data-licensing  
Published: 2026-09-29  
Updated: 2026-10-14

AI data licensing is a deal that lets an AI company use data you hold. In return you get a fee, a share of sales or a royalty.

> **Licensing means a company gets to use your records for a while, under rules.** It is like renting out a room you never use. You keep the house.

An outright transfer hands the data over once. A license pays each time the data is used, in [Troveo](/buyers/troveo)'s words. Before any of that, expect a buyer to ask for proof of your rights in every record.

The market holds four kinds of data, and only one is company records. Most prices for that kind come from the buyers themselves.

The full licensing process is in the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies).

## What does AI data licensing cover?

Four kinds of data change hands. Each has its own buyers and its own price evidence.

**Content libraries.** News, images, forum posts and video. Reuters reported in February 2024 that Google pays Reddit about $60M a year for its data ([The Decoder](https://the-decoder.com/reddit-signs-60-million-annual-training-data-deal-with-google/)).

Shutterstock's CEO said its AI licensing business made $104M in 2023 ([Bloomberg Law](https://news.bloomberglaw.com/artificial-intelligence/shutterstocks-ai-licensing-business-generated-104-million)). Yahoo Finance reported that Reddit's "other revenue," mainly data licensing, was $43M in the second quarter of 2026 ([Yahoo Finance](https://finance.yahoo.com/media-advertising/articles/reddits-data-licensing-revenue-grew-125000998.html)).

**Expert work.** Platforms such as [Mercor](/buyers/mercor) and Scale AI hire people to write, grade and correct model output. That is paid labor, so no company records change hands.

**Business content.** Video, audio and courses a company owns, such as a producer's archive, a training company's course library or a company podcast.

Troveo and [Protege](/buyers/protege) license video, and Bloomberg reported buyers pay about $1 to $4 a minute for it ([Bloomberg via The Star](https://www.thestar.com.my/tech/tech-news/2025/01/13/youtubers-are-selling-their-unused-video-footage-to-ai-companies)).

Our [guide to licensing your business's content](/guides/license-your-business-content-to-ai) covers consent from members, clients and guests.

**Company records.** Email, chat, tickets, CRM history and code. This is the newest kind, and public prices are few.

## How is a license different from an outright transfer?

Troveo puts it in one line: "A sale pays once. A license pays every time the data is used, and you keep ownership" ([Troveo](https://www.troveo.ai/resources/data-monetization-examples)).

micro1, Troveo and Protege use the words loosely. [micro1](/buyers/micro1) pays under an agreed data partnership and says companies keep ownership of their underlying data.

Troveo describes a revenue share on every deal, and it has not published the split. Protege describes revenue-share payments on qualified deals ([Protege](https://withprotege.ai/data-provider/video)).

[Pricing models](/answers/data-monetization-pricing) shows the figures for each. [License company data to AI](/answers/license-company-data-to-ai) lists what to settle in the contract.

## What does Epoch AI say about the supply of public text for AI?

Public text is a limited supply. Epoch AI estimated that the supply of good public human text could run out between 2026 and 2032 (80% interval) ([Epoch](https://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-data)).

For company records, the draw is a record of real work. micro1's CFO said in an undated recorded forum talk that the prize is the chain from a request to a finished deliverable.

One example, from micro1's David Revland, is a Slack message asking for a report, plus the report itself.

The demand case has one gap. Epoch's review of RL environments describes app replicas and lab partnerships with product companies, and does not discuss real company archives ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)).

## Who takes company records, and on what terms?

Direct programs include micro1, [Handshake AI](/buyers/handshake-ai) and Mercor's data arm. Troveo runs a marketplace. Google's $10M winning bid for Spirit Airlines' archive is pending court approval.

[AI data companies](/answers/ai-data-companies) sorts the players by lane. [Who buys company data for AI](/buyers) compares the buyers.

Stated figures come from the companies themselves. micro1's Enterprise Data Partnership listing says $100K-$2M+ for approved data packages. Handshake AI says $100K to $4M+. SimpleClosure reported nearly 100 shutdown deals at $10,000 to $100,000.

## What does an owner need first?

Three things, in order.

1. **A count.** List the systems you run, the years of records in each and the rough number of items. The ten-minute sheet is in [how much your business data is worth to AI](/guides/how-much-is-your-business-data-worth-to-ai).
2. **Rights.** Your email and chat hold other people's words, often under NDAs. Counsel flags trade secrets, legal privilege and employee talk that US labor law protects ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)). Run the [rights check](/rights-check).
3. **A plan to remove names.** [Removing names from data](/answers/data-de-identification) covers what each law test asks and who does the work in a deal.

Then take the free [fit check](/business). If you want help preparing the records, see the [managed readiness service](/managed-readiness).

Figure: comparison of Three kinds of data AI companies license: Content libraries (Many public deals), Video archives (Priced by the minute), Company records (emphasized: Newest kind, few public prices, Few public prices). So a price for company records is mostly the buyer's own claim.

## How should you read a company's own licensing claim?

As a claim. Troveo's [business-data page](https://www.troveo.ai/business-data) shows "$20M+" paid to licensors and "4x" average sales per dataset.

Troveo's number covers more than 7,000 rights holders across video, audio and other data. So it says little about business-data licensors ([Troveo press release](https://finance.yahoo.com/sectors/technology/articles/troveo-accelerates-ai-model-development-130500935.html)). The same page counts 40+ active buyers.

Calculators differ too, since each runs on its own assumptions about headcount and years of history. Ask what each number is built from, and who gets paid when you sign.

## What can go wrong?

Removing names is the weak joint. A label like "Person 14" keeps the links between records. The flight attendants' union in the Spirit case objected to a requirement to preserve "referential integrity" ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)).

The [risk map](/risks) lists the other failure points for data owners.

Check the buyer too. The [buyer directory](/buyers) and the [deals tracker](/deals) follow each player and each public figure.

Talk to a privacy or commercial lawyer in your state before you sign. Their view on rights counts for more than any buyer's estimate.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [The Decoder](https://the-decoder.com/reddit-signs-60-million-annual-training-data-deal-with-google/)
- [Bloomberg Law](https://news.bloomberglaw.com/artificial-intelligence/shutterstocks-ai-licensing-business-generated-104-million)
- [Yahoo Finance](https://finance.yahoo.com/media-advertising/articles/reddits-data-licensing-revenue-grew-125000998.html)
- [Bloomberg via The Star](https://www.thestar.com.my/tech/tech-news/2025/01/13/youtubers-are-selling-their-unused-video-footage-to-ai-companies)
- [Troveo](https://www.troveo.ai/resources/data-monetization-examples)
- [Protege](https://withprotege.ai/data-provider/video)
- [Epoch](https://epoch.ai/publications/will-we-run-out-of-data-limits-of-llm-scaling-based-on-human-generated-data)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [business-data page](https://www.troveo.ai/business-data)
- [Troveo press release](https://finance.yahoo.com/sectors/technology/articles/troveo-accelerates-ai-model-development-130500935.html)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)

---

# How Defined.ai uses your company data

> How Defined.ai handles supplier data: the consent proof it asks for, the license to share samples, the confidentiality terms, and what it doesn't say.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-defined-ai-uses-your-company-data  
Published: 2026-09-22  
Updated: 2026-10-14

[Defined.ai](/buyers/defined-ai) runs a marketplace where a business lists datasets that AI developers can license. It asks suppliers to prove consent and the right to license the data. It takes a license to share data samples with prospective customers. It publishes its [supplier terms](https://defined.ai/supplier-program).

We found no program built for a company's internal Slack and email archive.

This guide reads differently from the others. Defined.ai handles packaged datasets that a supplier has already cleared. The handling terms live in its supplier agreement more than in marketing copy.

Every claim below comes from Defined.ai's own pages, so treat it as self-reported. The product overview is in [the Defined AI data marketplace](/answers/defined-ai-data-marketplace).

> **Data licensing means a company gets to use your records for a while, under rules.** Defined.ai is the library's front desk. You hand in books that are already cleared, and it lends them out.

Our only referral relationship is with micro1, an AI data company. It may pay us a fee if a company we introduce goes ahead with it. You pay nothing extra. [How we are paid](https://vaultfish.com/disclosure)

## What data does Defined.ai ask for?

Its [data partner page](https://defined.ai/partnerships) says any business that generates or aggregates data can apply. It takes audio, video, image, text or mixed data.

Its [supplier terms](https://defined.ai/supplier-program) name four service types: data collection, crowd-as-a-service, evaluation of experience and reinforcement learning from human feedback. Its partnerships page says it wants content such as cinematic video, anime-style imagery, code and STEM material.

It does not list workplace chat, email or CRM records. A company archive holds your employees' and customers' words. That raises the consent question the marketplace is built around.

## How does Defined.ai collect it?

Suppliers apply and upload. The [partner page](https://defined.ai/partnerships) lists four reviews: account creation, a market review, a data quality review, and an ethics and compliance check.

The compliance step asks for proof that the data is ethically sourced. That means proper consent, paperwork and the right to license it for AI training.

There are no connectors and no exports run by Defined.ai. You package the data and you upload it.

Figure: steps Account and market review -> Consent and quality checks -> Dataset listed. Defined.ai checks consent proof before a dataset goes live on the marketplace.

## What does Defined.ai say it does to the data?

On supplier data, the pages we read don't say. The partnership page and the supplier terms don't mention anonymization or removing names and emails.

Its ethics page covers the company's own contributor programs. It says Defined.ai uses data minimization, pseudonymization, encryption and access controls throughout processing.

That describes data it collects from contributors. It is not a promise about a business's dataset.

So the burden sits with the supplier. If your dataset has people in it, you must show consent.

## Who receives it?

Customers. The [partner page](https://defined.ai/partnerships) says the marketplace serves startups and international AI heavyweights and names no one. The ethics page mentions clients in automotive, financial services, healthcare, gaming, robotics and retail.

One term deserves attention before you list. The [supplier terms](https://defined.ai/supplier-program) grant Defined.ai a worldwide, free, non-exclusive and unencumbered license to share data samples with customers so they can judge interest. Samples leave before any customer licenses the dataset.

## Does Defined.ai license your data or take it outright?

You license it. [The terms](https://defined.ai/supplier-program) describe the supplier relationship as non-exclusive, with no minimum commitment for business on either side. The partner page says usage follows defined licensing terms and agreed use cases.

It doesn't say who holds derived models or whether a customer can resell. We found no mention of exclusivity pricing.

## What about retention and deletion?

The pages don't set a retention period for datasets or a deletion process. The terms do set confidentiality. It runs for three years after the relationship ends. Materials go back within ten calendar days after termination.

That binds both sides on confidential information. It is not a promise to delete your dataset from customers' hands.

## What security does Defined.ai claim?

More than most buyers here. Its ethics page claims ISO/IEC 27001 and 27701 certification and an ISO/IEC 42001 certificate for AI management systems.

Its [partner page](https://defined.ai/partnerships) says suppliers go through legal, compliance and cybersecurity checks. We did not verify the certificate scopes.

The terms also require suppliers to follow Defined.ai's data protection policies as a non-negotiable condition. They let Defined.ai audit a supplier.

## Where does the risk sit?

On the supplier. Under [the terms](https://defined.ai/supplier-program), you agree to defend, indemnify and hold harmless Defined.ai and its affiliates against third-party claims. In plain words, you protect them from lawsuits.

Your liability under that clause is capped at two times the total value of the business conducted. The terms set no matching cap for Defined.ai and leave its side to a separate data license agreement.

Read [reps and warranties](/answers/what-reps-and-warranties-does-an-ai-data-buyer-ask-for) to see how these clauses work.

## How does Defined.ai pay?

Defined.ai pays through licensing fees from marketplace deals. The commission is not published, and the terms state no revenue split.

Defined.ai [reports](https://defined.ai/press-room/defined-ai-reports-sixty-five-percent-revenue-growth) 65% revenue growth in 2025. That is the company's own growth. Its [partnership page](https://defined.ai/partnerships) also quotes partner revenue, undated and unaudited. I leave those figures out, because they predict nothing for what your company would receive.

_Someone else's revenue is a bad number to plan a budget around._

## What should you ask Defined.ai?

- What is the commission, and when am I paid?
- Which data samples go to customers, and can I limit them?
- What happens to my dataset when I delist it?
- Are customers allowed to resell or build derived products?
- What consent evidence is enough for my kind of data?
- What does the 3-year confidentiality cover?

Run the [rights check](/rights-check) first, and see the other buyers in the [buyers directory](/buyers). Talk to a privacy or commercial lawyer in your state before you sign.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [supplier terms](https://defined.ai/supplier-program)
- [data partner page](https://defined.ai/partnerships)
- [reports](https://defined.ai/press-room/defined-ai-reports-sixty-five-percent-revenue-growth)

---

# How Troveo uses your company data

> How Troveo uses your company data: export-only collection, a documented scrubbing standard, confidential buyers, a license that keeps ownership, pay per deal.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-troveo-uses-your-company-data  
Published: 2026-09-21  
Updated: 2026-10-14

[Troveo](/buyers/troveo) says you export your company data with its guidance. It scrubs names and identifiers to one documented standard, then licenses the result to AI buyers for you. Per its page, ownership never transfers, and you share in the revenue on every deal.

Of the six buyers we read, Troveo asks for the least access to your systems. It also leaves the most open, because the standard is unpublished and the buyers are unnamed.

Everything below is Troveo describing itself. For the program overview, see [Troveo business data](/answers/troveo-business-data).

Our only referral relationship is with micro1, an AI data company. It may pay us a fee if a company we introduce goes ahead with it. You pay nothing extra. [How we are paid](https://vaultfish.com/disclosure)

## What data does Troveo ask for?

Troveo started with video and audio and added company data in 2026. Its [business-data page](https://www.troveo.ai/business-data) lists system groups with named examples:

- Chat and meetings, such as Slack, Teams, Google Chat and Zoom.
- Email and calendar, such as Gmail and Outlook.
- Storage, such as Google Drive, OneDrive, SharePoint and Dropbox.
- Work tools, such as Jira, Linear, Asana, Notion and Figma.
- Business systems, such as Salesforce, HubSpot, Zendesk, QuickBooks, Stripe, GitHub and Snowflake.

The page says buyers pay a premium for work they can follow across channels. It says more systems and more years of history raise the value.

## How does Troveo collect it?

By export. [The page](https://www.troveo.ai/business-data) says there is no OAuth into your systems and no platform to install. Troveo guides you through the export. Its stated process has four steps:

1. **Answer a few questions** You get an estimate.
2. **Agree standard terms**
3. **Names scrubbed, data packaged**
4. **Get paid**

You hold the files until you send them. That keeps your platforms' connection (API) terms out of the picture. It also leaves you responsible for what you put in the export.

## What does Troveo say it does to the data?

[It says](https://www.troveo.ai/business-data) it scrubs people's names, contact details, credentials and client identities under one documented standard. That happens before the data is put to work.

Its page shows a before and after. A sentence naming a customer and an email address becomes one with PERSON-014 and EMAIL-019 in their place.

The stand-ins in that example are consistent labels. That is pseudonymisation, and it keeps the trail across records. The same example also shows a file path replaced with a code.

The page does not show the standard itself. Nobody outside Troveo can check it. Ask for the written standard before you sign. Ask whether the mapping from labels to real names is kept or destroyed. See [pseudonymization versus anonymization](/answers/pseudonymization-vs-anonymization).

## Who receives it?

Unnamed buyers. [The page](https://www.troveo.ai/business-data) says your company is never named publicly. Buyers are introduced confidentially only after an agreement is signed. It mentions 40+ active buyers and describes them as a frontier AI lab, an AI startup and a Mag 7 company, without names.

That protects your privacy. It also means you cannot judge a buyer before you commit. Ask whether you get a veto over a named buyer, or at least a category list.

## Does Troveo license your data or take it outright?

It licenses. [The page's](https://www.troveo.ai/business-data) answer to who owns the data is "You do." Troveo licenses it on your behalf under terms you sign off on.

> **Data licensing means a company gets to use your records for a while, under rules.** It's like renting out a room you never use. You keep the house.

The page describes exclusivity as time-limited and something that has to be earned, with exit terms in the agreement. The license terms themselves are not published.

Figure: comparison of What Troveo says about your data: Ownership (Stays with you), Buyers (Unnamed until signing), Scrubbing (emphasized: Ask for it in writing, Standard unpublished). Ask for the buyer names and the scrubbing standard in writing.

## What does Troveo say about retention and deletion?

Troveo's business-data page does not address deletion. Troveo's [licensing guide](https://www.troveo.ai/resources/ai-data-licensing-deals) lists deletion at expiry among the terms to settle.

It also says models already trained survive a license ending, because nothing can be untrained. That is Troveo describing the industry in general.

What your agreement says is the only answer that counts. The wider picture is in [deleting after training](/answers/can-i-delete-my-data-after-an-ai-company-trains-on-it).

## What security does Troveo claim?

Little. The page describes privacy design. It publishes no certification, no encryption statement and no breach process.

The footer links a privacy policy and a support hub, and no security or terms page. We found no incident report. Ask for the security documentation you would expect from any vendor holding your records.

## How does Troveo pay?

[The page](https://www.troveo.ai/business-data) says you share in the revenue on every deal, the first and later ones, and that companies pay no fees. The split is not published.

It says typical full-company deals start at six figures. It shows an illustrative dataset licensed to several buyers, which is an example and not a real deal.

Troveo says it has paid more than $20 million to licensors. That figure covers all its licensing, not only business data, and it predicts nothing about what your company would receive. See [Troveo revenue share](/answers/troveo-revenue-share).

The page describes a repeat deal as a possibility. It promises you nothing.

## What should you ask Troveo?

- Can I read the scrubbing standard, and does it cover free text and file paths?
- Is the label-to-name mapping kept?
- Can I see buyer categories or veto a named buyer before a deal?
- What does the license let a buyer do with derived or enriched datasets?
- What is deleted at expiry, and what is not?
- What is the revenue split, and how often is it reported?

Run the [rights check](/rights-check) first, and see [is Troveo legit](/answers/is-troveo-legit) for the evidence on the company. Talk to a privacy or commercial lawyer in your state before you sign.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [business-data page](https://www.troveo.ai/business-data)
- [licensing guide](https://www.troveo.ai/resources/ai-data-licensing-deals)
- [Troveo: how much is company data worth](https://www.troveo.ai/resources/how-much-is-company-data-worth)
- [Troveo content owners](https://www.troveo.ai/content-owners)

---

# Business data licensing, and who buys it

> Business data licensing means letting AI buyers use your records, such as email, chat and tickets. What they look like, who buys and how a deal runs.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/business-data-licensing  
Published: 2026-09-16  
Updated: 2026-10-14

Business data licensing means letting a buyer, now mostly an AI company, use your company's operating records. Those records are email, chat, documents, tickets, CRM history, finance records and code.

> **Data licensing means a company gets to use your records for a while, under rules.** It's like renting out a room you never use. You keep the house.

[micro1](/buyers/micro1) runs a data partnership program. Its CFO said in an undated recorded forum talk that it looks for the chain of work from a request to a deliverable across tools. Other companies named in public sources include [Handshake AI](/buyers/handshake-ai), [Mercor](/buyers/mercor) and Google.

This guide is the company version of the topic. It covers what operating data looks like, who buys it, what a deal involves and what rules a business out. The full licensing process is in the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies).

Most payout figures below are self-reported.

## What counts as business data?

The companies list the same kinds of systems.

[Troveo](/buyers/troveo)'s page names these groups: chat and messaging, email and calendar, documents and files, work and knowledge tools, CRM, sales and support, business platforms, finance and operations, and data and code. It says internal custom systems can matter too ([Troveo](https://www.troveo.ai/business-data)).

micro1 names SOPs (written work procedures), knowledge bases, CRM data and project histories ([micro1](https://www.micro1.ai/data-partnerships)).

Buyers care most about the record of work. A single message matters less. Your tools hold that record in pieces, so the value sits in the trail across them.

## What does a valuable business record look like?

micro1's CFO gave one example in an undated recorded forum talk. A team member asks for a marketing budget. The record shows the request, the analysis in a spreadsheet, the back-and-forth and the approval across Slack, email and Google Workspace.

The CFO said the value is the path from request to approval across tools. The names and vendor line items matter less (self-reported).

Fortune calls such a recorded sequence of steps a trajectory ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)). A support ticket works the same way. It holds a problem, the steps and the resolution.

_A trajectory is a paper trail with a fancier name._

## Who buys it?

- **Google** placed the $10M winning bid for Spirit Airlines' archive, pending court approval. The archive holds about 100 million emails, 500 million Teams chats and 30 million lines of code. The sale was still before the bankruptcy court ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)).
- **Handshake AI** says partnerships range from $100K to $4M+.
- **Mercor** has a data arm with connectors to tools like Slack and Gmail. It has approached shuttered startups with offers reported up to $300,000 ([BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)).
- **Troveo** is a marketplace that licenses to buyers it rarely names.

[Who buys company data for AI](/buyers) has the sources. [AI data companies](/answers/ai-data-companies) sorts the players by lane.

## Where does micro1 fit?

[micro1](/buyers/micro1) runs company-data partnerships for AI development, with compensation under the agreed data partnership. Its Enterprise Data Partnership listing says $100K-$2M+ for approved data packages (self-reported).

## What does a deal involve?

David Revland, who works on human data at micro1, walked through its corpus process in an undated recorded forum talk (self-reported). It starts with an introduction, a first call and a data manifest form. The form asks which services you use and rough sizes per tool.

Then come a payment proposal, contract language, a transfer micro1's engineers help run, and payment after micro1 verifies the data. The CFO estimated, in an undated talk, about 10 hours over a week or two from one or two of your IT people.

Other buyers differ. Troveo says it needs no OAuth access (login permission) to your systems. Mercor says its platform pulls data through connectors and removes names.

Figure: steps Intro and first call -> Manifest form and contract -> Transfer, check, payment. micro1 pays only after it has checked the data.

## What does a business need to qualify?

The stated minimums differ by company. micro1's current public criteria list 30+ employees, mature operations and primarily English documentation, with US companies prioritized. Acceptance depends on its review and current demand.

Handshake AI's referral page says 20 or more full-time employees and 3 or more years in operation.

Firms whose records are mostly client files, such as law, accounting and healthcare practices, face the hardest rights questions. Much of what they hold is client material, and counsel decides what the firm can license.

## What can stop a deal?

Rights come first. Your email and chat hold other companies' information, often under NDAs. Counsel also flags trade secrets, privilege and employee talk that US labor law protects ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)). Your tools have terms too.

Salesforce's 2025 terms for the Slack connection tools use to pull data bar bulk export, persistent archives and use of that data in AI language models ([Hunton](https://www.hunton.com/insights/legal/salesforce-locks-down-slack-data-time-to-review-your-slack-api-terms)).

Removing names is the other weak joint. In an undated recorded forum talk, Revland said micro1 swaps names for consistent stand-ins, and micro1 staff said companies cannot scrub first. [Data de-identification](/answers/data-de-identification) covers what the legal tests ask.

## Where does the money question stand?

No public price list exists. Every public deal is a distressed company or a buyer headline. [Pricing models](/answers/data-monetization-pricing) shows the figures. [How much your business data is worth to AI](/guides/how-much-is-your-business-data-worth-to-ai) covers what moves a number.

## Where to start this week

Write one block per system: the tool, the oldest record, the approximate count, whose words are in it and any restrictions. Run the [rights check](/rights-check), then take the free [fit check](/business).

The [buyer directory](/buyers) and the [deals tracker](/deals) cover the players and the public figures. If you want help preparing the records, see the [managed readiness service](/managed-readiness).

Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Troveo](https://www.troveo.ai/business-data)
- [micro1](https://www.micro1.ai/data-partnerships)
- [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Hunton](https://www.hunton.com/insights/legal/salesforce-locks-down-slack-data-time-to-review-your-slack-api-terms)

---

# How do you monetize company data?

> How to monetize company data: improve your work, wrap it into products or package it. Ranked by evidence, with dated AI-buyer prices and rights checks.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/monetize-company-data  
Published: 2026-09-15  
Updated: 2026-10-14

You can monetize company data in three ways. Use it to improve how you work, wrap it into your products, or package information built from it as a product.

MIT research found that on average 51% of the returns came from improving work, 31% from wrapping and 18% from selling ([MIT Sloan](https://mitsloan.mit.edu/press/data-everybodys-business-fundamentals-data-monetization)).

Licensing records to an AI buyer is one of the newest kinds of data monetization. Its public prices are thin and mostly self-reported.

Whichever route you pick, ask first: do you have the right to use the records that way?

## What does it mean to earn from company data?

It means turning data you already hold into money or savings. The MIT researchers call the three ways improving, wrapping and selling.

Improving is a stocked shelf or a faster ticket queue. Wrapping is a recommendation that helps a customer buy. In their terms, selling is information that solves someone else's problem.

Those figures are averages across the firms studied, not a forecast for yours. They do show where to look first.

> **Data licensing is one way to earn from records. A company gets to use them for a while, under rules.** Think of renting out a room you never use. You collect rent and keep the house.

## Which routes can earn from data you already hold?

Here are five routes, ranked by how much public evidence supports each. The ranking is about evidence. It says nothing about payout.

1. **Improve your own work.** The best-evidenced route here, at 51% of returns in the MIT study. It needs your data and your team and no buyer. It is the least glamorous finding on this page and the best supported.
2. **Wrap data into your product.** 31% of returns in the same study. Think recommendations, benchmarks or alerts inside the products you already offer.
3. **Package information products.** 18% of returns. The MIT authors describe selling insights that solve problems, which goes beyond raw datasets.
4. **License records through a marketplace.** [Troveo](/buyers/troveo) says it matches you with buyers and removes names from the data to a documented standard. It pays a revenue share on every deal, with no platform fee for companies. It publishes no split ([Troveo](https://www.troveo.ai/business-data)).
5. **Join a direct data program.** [micro1](/buyers/micro1) and Mercor run direct programs. The evidence is the thinnest of the five. The labs at the end go unnamed, and the figures come from the programs themselves.

_The dullest route has the best evidence. The glamorous one has the best press._

## What do the public figures show?

There is no rate card. Troveo says so itself, and micro1 sets payment deal by deal. These are the public figures, each with its source type.

| Evidence | Figure | Source type |
|---|---|---|
| Google, Spirit Airlines archive, auction | $10M winning bid | Reported ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)) |
| micro1, Spirit Airlines offer | $12.5M | Reported ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)) |
| SimpleClosure, shut-down startups, about 100 deals in a year | $10,000 to $100,000 each | Reported ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)) |
| micro1, stated tiers | $100K+, $500K+, $1M+ | Self-reported ([micro1](https://www.micro1.ai/data-partnerships)) |
| micro1, approved data packages | $100K-$2M+ | Self-reported ([micro1's Enterprise Data Partnership listing](https://www.micro1.ai/company-referral)) |

The Spirit sale is one bankruptcy sale of one company's data, and the court has not approved it. Read the [Spirit Airlines page](/answers/spirit-airlines-data-sale) for the status. Our guide on [what your data is worth](/guides/how-much-is-your-business-data-worth-to-ai) shows why none of these is a rate you can multiply by your headcount.

Epoch AI's review of RL environments describes app replicas and lab partnerships with product companies, and does not discuss real company archives ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)). Demand is plausible. Public prices don't document it yet.

## What can stop a data license?

Rights. Daniel Goldberg, a partner at Frankfurt Kurnit, lists these risks: competing ownership claims, privilege waiver, privacy-policy promises, California sale rules, protected employee communications and data permanence.

It says to leave out privileged material, HR files, payroll, health data and union communications entirely ([FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

California defines a sale of personal information as a transfer to a third party for "monetary or other valuable consideration" ([Civ. Code 1798.140(ad)](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)).

On my reading, a revenue share fits that text, so ask counsel whether your deal is a sale. Customer contracts can bar the use outright.

Run our [rights check](/rights-check) before a buyer call. Read the [risk map](/risks) for where objections have come from. Talk to a privacy or commercial lawyer in your state before you sign.

## What does a micro1 corpus deal look like?

David Revland, who works on human data at micro1, described a corpus deal in an undated recorded forum talk, from introduction to payment. We group it into three stages.

micro1's CFO estimated in the same talk that the data transfer takes about 10 hours over a week or two from one or two IT people.

The CFO also described a recurring workflows track, where employees grade an AI agent on real daily work, and said a company can do both.

Figure: steps Intro and first call -> Manifest form and contract -> Transfer, check, payment. Expect a call before paperwork, and paperwork before any data moves.

The manifest form asks which tools you use and rough sizes per tool. Revland said micro1 then checks the delivered data before it pays, and that a possible top-up about six months later covers data created since the first transfer.

## How do you start?

Start with a readiness picture. A buyer call can wait. The free [fit check](/business) scores your records and shows what would move the score, with no upload. The [data readiness checklist](/answers/data-readiness-checklist) lists the same inputs as a list.

Read the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies) for the full process. Read [what goes in a data licensing agreement](/guides/data-licensing-agreement) before you see a contract.

If you want help preparing the file, see [managed readiness](/managed-readiness).

Information products are one route of three. If your own team can use the records to run the business better, that route carries the most public evidence. The [data monetization glossary entry](/answers/what-is-data-monetization) has the short version.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [MIT Sloan](https://mitsloan.mit.edu/press/data-everybodys-business-fundamentals-data-monetization)
- [Troveo](https://www.troveo.ai/business-data)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/)
- [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [micro1](https://www.micro1.ai/data-partnerships)
- [micro1's Enterprise Data Partnership listing](https://www.micro1.ai/company-referral)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Civ. Code 1798.140(ad)](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)

---

# micro1 data partnership: how the program works

> The micro1 data partnership program, walked through: three ways in, the process from first call to payment, eligibility, personal info and pay.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/micro1-data-partnership  
Published: 2026-09-07  
Updated: 2026-10-14

A [micro1](/buyers/micro1) data partnership is a deal where micro1 pays a company under an agreed data partnership for use of its workplace records, or for employees' graded work with an AI agent.

Its [data partnerships page](https://www.micro1.ai/data-partnerships) lists tiers from $100K+ to $1M+. It describes three ways in: a corpus deal, a workflows track and a referral.

> **Data licensing means a company gets to use your records for a while, under rules.** Picture an attic full of old boxes. micro1 wants to read through them, and it pays you under a partnership agreement.

Every number in this guide comes from micro1 describing itself, on its pages or in an undated recorded forum talk. Read each as a claim, not a measured result.

We read micro1's pages and worked from the recording of a forum talk by its CFO and team, the third Company Data Partnerships session. The recording has no date and its captions are automatic, so a few figures carry a "verify" flag below.

## What does micro1 say it is?

In an undated recorded forum talk, micro1's team called the company a data lab. It trains and improves AI models and agents by supplying data in formats models learn from. Its company data page says it helps organisations turn operational data into revenue.

For the company profile and funding figures, see [what is micro1](/answers/what-is-micro1).

The pitch for company data is simple. Hired experts build synthetic training environments that imitate the real world. micro1's CFO said the best training environment is the real world itself.

A real marketing budget request shows who asked, which tools they used, which spreadsheets changed and who approved it. That chain from request to deliverable is the product. The names and vendor line items matter less.

## What are the three ways in?

**1. Corpus deal.** This is the Enterprise Data Partnerships track, a partnership covering a company's data corpus.

micro1 says companies keep ownership of their underlying data. Confirm licensed uses, duration, exclusivity and payment structure in the proposed agreement.

micro1's page lists $100K+, $500K+ and $1M+ tiers set per deal, and its Enterprise Data Partnership listing says $100K-$2M+ for approved data packages.

Its CFO said in an undated recorded forum talk that deals near $2 million are real. In the same talk, David Revland, who works on human data at micro1, said it may come back about six months later for a top-up covering data created since the first transfer.

**2. Workflows track.** Employees use a general-purpose AI agent on their real daily work and grade how it did. They also say how it should have done the task. micro1 receives the evaluation and the agent's trajectory.

In the undated talk, the CFO described pay per completed task in the high hundreds of dollars for a couple of hours of work. We could not confirm that figure word for word in the captions (verify).

The company chooses how many employees take part and how often. A company can run both tracks.

**3. Referral.** micro1 also pays people who introduce a company. If someone introduced you, ask them in writing whether micro1 pays them.

Figure: comparison of Three ways into micro1's data program: Corpus deal (emphasized: One transfer, no repeats, One transfer), Workflows (Repeats by task), Referral (An introduction). Most companies start with the corpus deal, and workflows can run alongside it.

## What is the process for a corpus deal?

micro1's David Revland described seven steps in the undated recorded forum talk.

1. A referral, a peer introduction or organic contact starts it.
2. A first call with a data partnership manager.
3. A data manifest form on micro1's data platform, as described in the undated talk.

   It asks which services you use (Google Workspace, Slack, project management tools, GitHub and so on) and rough sizes per tool. micro1 supplies instructions for finding sizes, for example in Google Drive.
4. micro1 sends a payment proposal.
5. Contract language.
6. The transfer, with micro1's forward-deployed engineering team helping your IT people. In the undated talk, the CFO estimated about 10 hours of work from one or two of your IT people over a week or two. [How long a transfer takes](/answers/how-long-does-a-data-transfer-take) has the detail.
7. micro1 verifies the delivered data and then pays.

Figure: steps Intro and first call -> Manifest form and contract -> Transfer, check, payment. Scope and contract come before any data leaves your systems.

The page version is shorter. It lists four stages: a simple assessment, an evaluation, a discovery and privacy discussion, and a partnership agreement. The talk's seven steps are the same path in more detail.

## Who qualifies?

micro1's current public criteria list 30+ employees, mature operations and primarily English documentation, with US companies prioritized. Acceptance depends on its review and current demand. Confirm exceptions directly with micro1.

- **Employees.** The data partnerships page and the referral page say 30 or more.
- **Operations.** Mature operations with established documentation, knowledge bases, CRM metadata and project histories.
- **Language.** Primarily English. The pages give no percentage.
- **Place.** US companies first, then other Western markets.
- **Volume.** The pages state no data-size threshold. A storage count still helps you describe your records.

The undated recorded talk gave earlier, different figures: at least 20 full-time employees, about 1 TB across the corpus and about 80% English. It also named any industry, a US and Canadian preference and no cap on the number of companies.

Those figures are history, and they do not decide who qualifies now. Every company goes through micro1's evaluation. The referral page says a referral does not guarantee acceptance.

## What happens to personal information?

Personal information here means names, emails and phone numbers. micro1's CFO said in the undated talk that it does not want any.

In the same undated talk the CFO said every corpus goes through a scrubbing step, which it samples at several points before use. The page says personal information "is removed when relevant."

Revland said micro1 swaps names instead of blacking them out. A name is replaced with a consistent stand-in across every platform. The same person keeps the same stand-in in Slack and in Google Workspace, so the role and cross-tool trail survive.

For that reason micro1 staff said in the undated talk that companies cannot pre-scrub. The same talk said removing names first would destroy the trail micro1 needs. micro1 refers to its own data trust principles.

Counsel and regulators treat swapped names with caution. The [UK ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/) says pseudonymised data is still personal data for anyone who holds the extra information. The union objection in the Spirit Airlines sale turned on the same worry (Time).

Ask micro1 who holds the mapping between real and stand-in names, and how long it keeps it.

The page also says data is held in isolated pipelines. It is kept for agreed periods and deleted on request or at the end of the engagement, in line with the contract. We found no security certification published by micro1.

## What should you ask before the first call?

- Which of my systems does the manifest cover, and which stay out?
- Who sees the data besides micro1, and does the contract name the AI labs that receive it?
- Who holds the real-to-stand-in name mapping, and for how long?
- What does the contract say about my warranties, the promise to protect micro1 from lawsuits (the indemnity) and any cap on micro1's liability? [Reps and warranties](/answers/what-reps-and-warranties-does-an-ai-data-buyer-ask-for) shows the published pattern.
- If micro1 is breached, who pays and who notifies? See [who is liable after a breach](/answers/who-is-liable-if-the-data-buyer-is-hacked).
- Is the payout one-time, and is a top-up possible?

Before any call, run the [rights check](/rights-check) on the systems you plan to include. Customer contracts, employee conversations and tool terms such as Slack's API terms all sit between you and a clean deal.

The [risks page](/risks) lists what the companies publish about security and removing names. Want a person to organise the facts? [Managed readiness](/managed-readiness) is the service we run. We never take custody of your data.

## What does micro1's own video add?

A micro1 referral partner spoke in a micro1 video. The publish date is not stated, the video is promotional, and she earns from referrals. Her statements differ from the undated recorded talk in places, so both are printed here.

- **Payout clock.** Payouts typically arrive about 60 days after the company engages with micro1, with "engaged" undefined. The undated recorded talk and micro1's pages give no day count (about 2:54).
- **Form and payment proposal.** The form takes under 10 minutes and a payment proposal follows within a few days (about 2:20 and 2:34).
- **License or sale.** She calls it a license in which the company keeps full access. In the earlier, undated talk the CFO called the corpus track a one-time sale (about 12:29). micro1's current page says companies keep ownership of their underlying data.
- **Removing names.** She says it removes names, emails and phone numbers. Revland described a consistent stand-in swap (about 13:27).
- **IT effort.** She says no IT team is needed. In the undated talk the CFO said about 10 hours from one or two IT people (about 12:12).

_Two tellings of one program. The signed contract is the one that counts._

## What does micro1 say you gain?

Two things.

First, a new revenue stream from data you already store. Second, for the workflows track, a team that works with AI agents more fluently, "more AI native" in micro1's phrase.

micro1 reports a large business, and its pages name no lab clients. Weigh both claims with that in mind.

Compare it with the other published routes in the [buyers directory](/buyers) and the [deals tracker](/deals). Read [how to license company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies) for the whole market. Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [data partnerships page](https://www.micro1.ai/data-partnerships)
- [UK ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)
- [micro1 company referral](https://www.micro1.ai/company-referral)
- [micro1 referral terms](https://www.micro1.ai/legal/referral-terms)

---

# How Mercor uses your company data

> How Mercor uses your company data: the connectors it asks for, the automated name removal, who it says sees it, pay, and the March 2026 breach.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-mercor-uses-your-company-data  
Published: 2026-08-28  
Updated: 2026-10-14

[Mercor](/buyers/mercor) says it connects to your work accounts and pulls out the data. It says an automated pipeline removes names and details.

Then Mercor licenses the result to AI labs. It says no lab or outside vendor sees your raw data. It pays by wire after the anonymized dataset is delivered.

> **Data licensing means a company gets to use your records for a while, under rules. You keep ownership.** Think of renting a room you never use. The guest stays under house rules, and you keep the house.

All of that comes from [Mercor's own data page](https://www.mercor.com/data/). Mercor also had a security incident in March 2026 that exposed contractor data. We cover it below. For the pricing and offers around the program, see [Mercor data licensing](/answers/mercor-data-licensing).

Our only referral relationship is with micro1, an AI data company. It may pay us a fee if a company we introduce goes ahead with it. You pay nothing extra. [How we are paid](https://vaultfish.com/disclosure)

## What data does Mercor ask for?

Workflow data from the tools a company uses daily. [Mercor's data page](https://www.mercor.com/data/) names Slack, Microsoft Teams, Gmail, Confluence, Salesforce, Outlook, GitHub, Google Drive and Notion. It says it supports 100+ tools.

It also lists target industries with example records:

- Legal contracts and case files
- Healthcare records
- Financial statements
- Manufacturing logs
- Real estate documents
- Supply chain manifests

Read that list twice. Healthcare records, case files and financial statements carry privacy and confidentiality duties of their own. Those are a question for your lawyer, not Mercor.

## How does Mercor collect it?

Through connected accounts. The page describes four stages: scope, connect, process and payout. In the first, you choose what moves. The page says no data leaves your systems until you approve it. Then you link accounts. The page says credentials are held per session and never shared between connections.

Connecting accounts means your platforms' own API terms apply. An API is a connection between two systems. The buyer can read whatever the connection scope allows. That is a different risk from handing over an export file.

Troveo says it doesn't connect to your systems at all. See [the Mercor breach page](/answers/mercor-data-breach) for why the difference matters.

## What does Mercor say it does to the data?

[Its page](https://www.mercor.com/data/) says the platform "extracts and de-identifies" your data into an asset. It says the extraction and anonymization are fully automated. It doesn't say what that means in practice. It names no standard, no method and no sampling. Whether names are removed or swapped is not stated. Neither is whether free text is checked.

Ask for the specification in writing. A swap that keeps each person's stand-in consistent behaves differently from removal when someone tries to re-identify the records.

## Who receives it?

The page says the dataset goes to "frontier AI" labs and names none. It says no AI lab buyer and no third-party vendor ever sees your raw data. Mercor's engineering team oversees the pipeline but does not manually review your data. All access is logged under strict internal controls.

The only named buyer-side facts are bids on one archive. Mercor bid $7.5 million for Spirit Airlines' archive, and Google won the auction at $10 million, pending court approval, per [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/) ([the Spirit sale timeline](/answers/spirit-airlines-data-sale)).

Mercor also reportedly offered up to $300,000 to shuttered startups for chat, drive and email archives, per [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d).

## Does Mercor license your data or take it outright?

[The page](https://www.mercor.com/data/) says "We license it, package it, and pay you for it." It publishes no contract. So it says nothing on exclusivity, term or what the license lets a lab do with derived data. It names no retention period and no deletion process. On the points that matter after signing, Mercor doesn't say.

Figure: steps You connect accounts -> Mercor says it removes names -> Wire after delivery. Mercor says its pipeline runs without a person reading your raw data.

## What security does Mercor claim?

Four things, per [its data page](https://www.mercor.com/data/): SOC 2 Type II, encryption in transit, encryption keys managed in AWS KMS, and an audit log of every access. We could not verify the audit's scope, the auditor or the test period.

Then the incident. [Mercor said](https://www.mercor.com/blog/update-on-mercor-security-incident/) a poisoned release of the open-source LiteLLM library, live for a short window in March 2026, affected it. Mercor said the same release affected a significant number of other companies. [The Next Web](https://thenextweb.com/news/meta-mercor-breach-ai-training-secrets-risk) reported more than 40,000 current and former contractors and customers and about 4 TB of material. That included source code, a user database, video interviews and identity documents.

[Mercor says](https://www.mercor.com/blog/update-on-mercor-security-incident/) the impact on customer information was very limited. A class action, *Ananthula v. Mercor.io Corp.*, was filed on 21 April 2026 in the Northern District of California. Its claims are allegations. No company-archive data was reported exposed.

A buyer that holds many companies' records in one pipeline is a target. Its certificate is a claim to check. Ask for the SOC 2 report itself, read the scope and the dates, and ask who the auditor was.

_A badge on a website is a nice start. The report behind it is the real thing._

## How does Mercor pay?

Mercor says pay depends on data volume, the tools you connect and the depth of the data. Connecting more tools generally pays more. [Its data page](https://www.mercor.com/data/) says payment is a wire "typically within 2-4 weeks" of the signed agreement, issued on delivery of the dataset to the buyer. It says it covers extraction, anonymization and transfer costs and takes no setup fee or deduction. No price list is published. Nothing here promises you a payout.

## What should you ask Mercor?

- What connection scope do you need for each tool, and can it be read-only?
- What does the name removal do to names, emails and free text?
- How long is raw and processed data kept, and who deletes it?
- Which labs can receive the dataset, and can they resell it?
- Where is the SOC 2 report, and what does it cover?
- If a breach touches my records, who notifies and who pays? See [who is liable if the buyer is hacked](/answers/who-is-liable-if-the-data-buyer-is-hacked).

The evidence on the company is in [is Mercor legit](/answers/is-mercor-legit). The [rights check](/rights-check) comes before any account connects. Talk to a privacy or commercial lawyer in your state before you sign.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Mercor's own data page](https://www.mercor.com/data/)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [BigGo Finance](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)
- [Mercor said](https://www.mercor.com/blog/update-on-mercor-security-incident/)
- [The Next Web](https://thenextweb.com/news/meta-mercor-breach-ai-training-secrets-risk)
- [CourtListener docket, Deboni v. Mercor.io Corp.](https://www.courtlistener.com/docket/73129103/deboni-v-mercorio-corporation/)

---

# How to choose an AI data broker

> How to choose an AI data broker: eight checks on entity, terms, numbers, buyers, fees, data handling and contract, with dated examples from public buyer pages.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-to-choose-an-ai-data-broker  
Published: 2026-08-28  
Updated: 2026-10-14

Pick the buyer that lets you check the most. Seven checks cover most of it before you send a single record, and each takes minutes. In order:

1. A named legal entity
2. Live terms and privacy pages
3. Numbers that agree with each other
4. Named buyers
5. A published fee
6. A clear rule for when your data moves
7. A contract that sets liability

Each check below uses examples from public buyer pages.

Nobody has outside proof of what any buyer paid a business for its data. So these checks test whether a firm's story hangs together and how much it shows.

You make the call on each firm. The firms report their own figures, and none is a promise of what you would receive.

> **Data licensing means a company gets to use your records for a while, under rules. You keep ownership.** Think of renting a room. Before you hand over a key, you check who the landlord is, what the lease says and what it costs.

Figure: steps Check the entity -> Test the numbers -> Read the contract. Run all three before you sign anything.

People search for "data broker" when they mean an AI data company. Legally, a data broker resells consumer data and must register in some states. This site calls AI data companies buyers.

## Check 1: does the buyer name a legal entity?

Look at the footer and the terms. A named company gives you someone to hold to an agreement.

[SimpleClosure](/buyers/simpleclosure)'s terms name "Simple Closure Inc." Some referral sites give only a trading name, or an operator with no address or registration.

[Troveo](/buyers/troveo)'s business-data page footer shows "Troveo 2026" and no legal entity. Ask any of them for the contracting entity before a call.

## Check 2: are the terms and privacy pages live?

Open both pages and check the date. SimpleClosure's terms cover its Asset Hub. Troveo's `/terms` and `/privacy` paths returned 404. A missing page can be a site fault. Ask for the supplier agreement in writing.

## Check 3: does the headline match the calculator?

Most buyers show an estimator. Test its top end against the headline.

[Handshake AI](/buyers/handshake-ai)'s calculator example (100 employees, 10 years) gives $383,083 to $896,969. Its headline says $100K to $4M+. It calls the example "indicative only."

When a calculator tops out far below the headline, ask which companies reach the headline.

## Check 4: do the buyer's own pages agree?

Compare totals across its pages. Troveo shows $20M+ paid on its business-data page. Its content-owners page shows "$50M+" with no date.

Handshake AI shows $250M paid to partners on Handshake AI's [company referral page](https://joinhandshake.com/ai/data-partnerships/referral) and $500M+ in payouts on its fellowship page. These are company-wide totals, and none of them predicts what your company would receive.

Each pair may cover different groups. The pages don't say. Ask which group each number counts.

## Check 5: does it name who gets the data?

Marketplaces describe their buyers as a "frontier AI lab" or a "selected partner." Troveo's page uses "Mag 7 company." The Spirit Airlines bids from Google, [Mercor](/buyers/mercor) and [micro1](/buyers/micro1) were public only because a bankruptcy court ran the auction ([Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)).

Ask which kinds of buyer will see your data, and under what contract.

A micro1 referral partner made two points in a micro1 video (about 17:00 and 22:31). Owners are right to check that the company is reputable and that other companies are signing.

And the first question is who micro1 is. Ask who else is signing, and which lab ends up with the data.

## Check 6: are the fees and splits published?

Troveo shares revenue "on every sale" and charges companies no fees. It doesn't publish its split. An advisor paid on success should state its percentage before you sign.

SimpleClosure's terms point to fees in a separate order form, and its AssetHub page says it takes an unpublished percentage on a completed deal.

micro1 and Handshake AI publish tiers or ranges for what they pay. That gives you a figure to test. But a price range doesn't show what a middleman keeps. Get the percentage in writing.

## Check 7: when does your data move, and how is it prepared?

The companies differ here.

- [Troveo](https://www.troveo.ai/business-data) says it needs no OAuth access to your systems and no platform to install.
- Mercor connects to your tools and pulls the data through connectors.

micro1 staff said in an undated recorded forum talk that companies can't scrub first and micro1 must change the data itself. micro1's CFO said its engineers help your IT people with the transfer.

Mercor says no lab buyer or vendor sees raw data. Its own March 2026 incident update says a supply chain attack on LiteLLM affected it.

Whatever the answer, ask who holds the data, for how long and where it is deleted.

## Check 8: what does the contract say?

Daniel Goldberg, a partner at Frankfurt Kurnit, advises that a purchase agreement should "restrict how the buyer can use the data, allocate privacy risk, and address what happens if something goes wrong" ([Goldberg](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

Ask about four things: the promises you must give, who pays if the buyer is breached, deletion on request, and exclusivity. Epoch's interviews put exclusive RL-environment deals at roughly 4x to 5x the non-exclusive price. Exclusivity changes value ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)).

## What about domain age and reviews?

Domain age is a weak signal. A domain registered this year leaves no track record to read. Mercor's domain, though, predates the company by two decades. Treat age as a cue to ask for references.

Reviews mostly describe worker platforms. We found none from companies that licensed business data. Ask each buyer whether a past licensor will talk to you.

A micro1 referral partner said in a micro1 video that she connects hesitant owners with owners who finished (about 19:34). Ask a reference four things:

- Who ran the export?
- How long did it take?
- What was removed?
- When did payment arrive?

_Domain age is a weak clue. A brand-new site can be honest, and an old one can still say "TBD" about fees._

## What should you do next?

Run the [rights check](/rights-check) and read the [risk map](/risks). Then bring the answers to a first call. The [buyer directory](/buyers) and [best AI data brokers](/guides/best-ai-data-brokers) list each firm's public terms. [AI data broker red flags](/answers/ai-data-broker-red-flags) ranks the patterns above.

For help getting your records and rights ready before any buyer sees them, see [managed readiness](/managed-readiness). We work from descriptions and counts, never your records.

Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [company referral page](https://joinhandshake.com/ai/data-partnerships/referral)
- [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)
- [Troveo](https://www.troveo.ai/business-data)
- [Goldberg](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)

---

# How SimpleClosure uses your company data

> How SimpleClosure uses your company data through Asset Hub: what it takes, the perpetual license in its terms, the consent duty on you, and what it skips.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-simpleclosure-uses-your-company-data  
Published: 2026-08-17  
Updated: 2026-10-14

[SimpleClosure](/buyers/simpleclosure)'s Asset Hub sells code and workspace data from shutting-down startups to AI buyers. [Its terms](https://simpleclosure.com/tos/) take a perpetual, irrevocable license to what you submit and make you hold consents from the people in it. [Gizmodo reported](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916), citing Forbes, nearly 100 deals at $10,000 to $100,000 per company.

SimpleClosure says it handles collection, strips personal details and sends the data to the buyer. Its terms are published in full, so it is the easiest buyer here to check.

It was built for a company that is closing. Its [AssetHub page](https://simpleclosure.com/product/assethub/) now says operating companies can license too. The [buyers directory](/buyers) lists other options. For the product itself, see [SimpleClosure Asset Hub](/answers/simpleclosure-asset-hub).

Our only referral relationship is with micro1, an AI data company. It may pay us a fee if a company we introduce goes ahead with it. You pay nothing extra. [How we are paid](https://vaultfish.com/disclosure)

> **Data licensing means a company gets to use your records under rules. You still own them.** Think of renting a room. Most renters get a lease with an end date. This lease has none, and you can't cancel it.

## What data does SimpleClosure ask for?

SimpleClosure's [terms](https://simpleclosure.com/tos/) call what you submit "digital assets." That covers software codebases, source code repositories and technical documentation. It also covers workplace data exports, such as emails, messaging exports and project management records.

The [launch release](https://finance.yahoo.com/sectors/technology/articles/simpleclosure-launches-asset-hub-unlock-130000668.html) sorts these into three groups:

- Code is live.
- Workspace data, such as email, Slack, Notion and Google Drive, is in beta.
- Domains, patents, trademarks and equipment are in development.

## How does SimpleClosure collect it?

Through its services. The release says SimpleClosure manages the deal end to end, including data collection. The terms say you submit assets through the services. For code, intake starts with a connection to GitHub.

SimpleClosure shares no more detail on how it pulls workspace data. Ask whether you run an export or it runs a connection.

## What does SimpleClosure say it does to the data?

The [launch release](https://finance.yahoo.com/sectors/technology/articles/simpleclosure-launches-asset-hub-unlock-130000668.html) says SimpleClosure removes personal information before transfer. It means names, emails and other details that point to a person.

[The terms](https://simpleclosure.com/tos/) matter more here. You promise that you have all required consents from the people in the assets. That includes employees, contractors, customers, investors and vendors. The terms name the CCPA, CPRA and GDPR. You must also disclose open-source parts in the assets.

So the process says SimpleClosure scrubs. The contract says you promised the data was fine to share. Both can be true. If the second turns out false, you answer for it.

The terms promise no scrubbing. They say only that its evaluation may include an assessment of embedded personal information.

## Who receives it?

The terms list three groups: SimpleClosure, its affiliates and subsidiaries, and third-party buyers it introduces. The release names the buyer types as AI labs, reinforcement learning researchers and enterprise AI companies. It names no buyers.

[Gizmodo reported](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916), citing Forbes, nearly 100 deals at $10,000 to $100,000 per company.

## Does SimpleClosure license your data or take it outright?

You license it. The [license](https://simpleclosure.com/tos/) is the broadest on our list. It is non-exclusive, transferable, sub-licensable, royalty-free, fully paid up, perpetual, irrevocable and worldwide.

It lets SimpleClosure use, copy, change, share, license or sell the assets for any lawful purpose. That includes licensing or selling them to third parties. You are owed no further consent or pay, except what a separate asset assignment or purchase agreement says.

Figure: comparison of SimpleClosure's license on your assets: Term (Perpetual), Revocation (emphasized: You cannot take it back, Irrevocable), Resale (Sub-licensable). Once granted, this license stays with SimpleClosure and its buyers.

The terms state no revenue share and no fee. Payment sits in a separate agreement. The price of the license is whatever that agreement says.

Its [AssetHub page](https://simpleclosure.com/product/assethub/) adds that there is no upfront cost and SimpleClosure takes a percentage only on a completed deal. The percentage is not published.

## What about retention and deletion?

The terms set no automatic deletion. After the contract ends, [the terms](https://simpleclosure.com/tos/) say SimpleClosure provides your data on request for 30 days.

That returns your copy. It says nothing about deleting what a buyer holds, and the license can't be revoked. See [deleting after training](/answers/can-i-delete-my-data-after-an-ai-company-trains-on-it).

## What security does SimpleClosure claim?

The terms don't cover it. We found no certification or encryption statement in the material we read. A closing company's email archive is sensitive. Raise this gap first.

## Where does the risk sit?

[The terms](https://simpleclosure.com/tos/) make you defend and protect SimpleClosure from lawsuits over four things:

- Breaches of your promises
- IP claims
- Privacy violations
- Open-source non-compliance

SimpleClosure caps its own liability at the greater of what you paid in the prior six months or $100. That is a lopsided pair. You can owe a lot. It likely owes $100 at most. [Reps and warranties](/answers/what-reps-and-warranties-does-an-ai-data-buyer-ask-for) explains what such promises cost.

## How does SimpleClosure pay?

It doesn't publish the fee or the split. The only figure is the reported range of $10,000 to $100,000 per company. That figure is second-hand and promises you nothing. See [how much SimpleClosure costs](/answers/how-much-does-simpleclosure-cost).

_An irrevocable license is the only lease where the landlord never has to give the key back._

## What should you ask SimpleClosure?

- What is the fee, and is it taken before or after I'm paid?
- Can I leave out messages from people who haven't agreed to share them?
- Who removes the personal details, and with what method?
- Can the irrevocable license cover training uses only?
- What happens to my data if the buyer resells it?
- Which promises can I limit?

Start with the [rights check](/rights-check). Then read [is SimpleClosure legit](/answers/is-simpleclosure-legit). Talk to a privacy or commercial lawyer in your state before you sign.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Its terms](https://simpleclosure.com/tos/)
- [Gizmodo reported](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [AssetHub page](https://simpleclosure.com/product/assethub/)
- [launch release](https://finance.yahoo.com/sectors/technology/articles/simpleclosure-launches-asset-hub-unlock-130000668.html)
- [SimpleClosure Asset Hub announcement](https://simpleclosure.com/blog/insights/simpleclosure-asset-hub/)
- [SimpleClosure CEO post](https://simpleclosure.com/blog/guides/how-real-are-those-million-dollar-data-offers/)

---

# How to prepare company data for AI licensing

> How to prepare company data for AI licensing: a five-item packet, what intake forms ask first, who does the scrubbing, and the rights step.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-to-prepare-company-data-for-ai-licensing  
Published: 2026-08-13  
Updated: 2026-10-14

Build a one-page packet before any first call, and leave the scrubbing to the company you license to. The packet lists:

- The systems you run
- The oldest record and a rough size in each
- Whose words sit in them
- What stays out
- Who can run the exports

Why not scrub first? [micro1](/buyers/micro1) staff said in an undated recorded forum talk that companies can't pre-scrub. micro1's method swaps names for matching stand-ins across tools and needs the original links intact.

Preparing here means gathering facts and checking rights. It does not mean data engineering. This guide covers the packet, what intake forms ask first, the scrubbing question and the rights step.

Your counsel owns the rights step. Most facts below come from micro1, Troveo, Mercor and Handshake AI themselves.

> **Data licensing means a company gets to use your records for a while, under rules. The contract says who keeps ownership.** Think of clearing out an attic before a guest stays. You list what's up there and what stays locked. The guest does the dusting.

## Which five items go in the packet?

1. **Systems and owners.** List every tool where work gets recorded. That means email, chat, CRM, helpdesk, project boards, documents, call recordings and code. Next to each, write the name of the person who runs it and can pull an export.
2. **Dates.** Note the oldest record you still hold in each system, and any gaps.
3. **Rough size per tool.** micro1's data manifest form, as micro1's David Revland described it in an undated recorded forum talk, asks which services you use and a rough size for each.

   The talk gives steps for finding sizes, for example in Google Drive. Our page on [checking your Google Drive size](/answers/micro1-google-drive-data) shows where. Screenshot any total with the date visible.
4. **Whose words.** Staff, customers, partners, contractors and guests in shared channels. Note any contract, NDA or tool term that mentions secrecy, data use or AI.
5. **Exclusions.** List the categories that stay out of any deal. Your counsel signs off on them.

The [ten-minute check in the worth guide](/guides/how-much-is-your-business-data-worth-to-ai) has a template for the first four items. The [readiness checklist](/answers/data-readiness-checklist) lists the thresholds each program publishes.

Figure: steps List your systems -> Mark exclusions -> Counsel review. Your counsel signs off on what stays out.

## What do intake forms ask first?

Each company opens with a different slice of the packet.

| Company or marketplace | First ask | Source |
|---|---|---|
| micro1 | Which services you use and rough sizes per tool, on a data manifest form | Recorded forum talk, date not stated |
| Troveo | Data estimate details, with no access to your systems | [Business data page](https://www.troveo.ai/business-data) |
| Mercor | Which tools to share, from 100+ supported connectors | [Data page](https://www.mercor.com/data/) |

[Mercor](/buyers/mercor) connects to the tools you choose. [Troveo](/buyers/troveo) says you need no access to start. Keep raw data in your own systems until a signed agreement says who receives it.

Our page on [the data manifest form](/answers/what-is-a-data-manifest-form) explains micro1's version.

## Should you scrub the data first?

The companies say the cleaning is their job.

Revland said in an undated recorded forum talk that micro1 swaps each person for the same stand-in in Slack and in Google Workspace. That way the role and the trail across tools survive.

micro1 staff said in the same talk that redacting first destroys that trail. [Mercor](https://www.mercor.com/data/) says extraction and anonymization are automated and it covers the cost.

[Handshake AI](https://joinhandshake.com/ai/data-partnerships/) says it removes direct and indirect identifiers before delivery. The partner stays responsible for the legal side of its transfer.

[Troveo](https://www.troveo.ai/business-data) says it scrubs names, contact details, credentials and client identities under one documented standard.

Removing names from emails and chats is hard. Counsel warns that text with the names removed can still reveal sensitive details about people or small groups ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)). Our pages on [removing names from data](/answers/data-de-identification) and [micro1's name swapping](/answers/micro1-pii-transformation) go further. The [anonymization guide](/guides/data-anonymization) covers the legal tests.

What you can do early is decide what stays out of scope. Leaving something out costs less than scrubbing it.

## What does the rights step involve?

Your counsel owns this step. We judge no contract. Daniel Goldberg, a partner at [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes), listed what a company reviews before a data deal:

- Whether you own or control the data. Email and chat hold third-party information, often under NDAs.
- Trade secrets and privileged messages.
- Your privacy policy and employee handbook, and [the FTC's view](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive) of quiet retroactive changes.
- Privacy laws such as the CCPA, and sector rules for health or financial records.
- Section 7 of the [National Labor Relations Act](https://www.nlrb.gov/about-nlrb/rights-we-protect/the-law/interfering-with-employee-rights-section-7-8a1). It can protect employee talk about pay and conditions.
- The limits of removing names.

His process advice has four parts. Gather legal, privacy, security, HR and communications people. List the sources. Confirm ownership through vendor contracts. Define exclusions and write down the review.

Your packet is the raw material for that review. The [rights check](/rights-check) asks seven questions on the same ground. [What data you can't license to AI](/answers/what-data-can-i-not-sell-to-ai) lists the categories that usually come out. The [risk map](/risks) shows where deals break.

_A packet with a clear "no" column beats a packet that says yes to everything. Lawyers like it, too._

## Who needs to be involved on your side?

You need one person with admin rights per system. Add whoever speaks for legal and HR. micro1's CFO estimated, in an undated talk, the transfer at about 10 hours of work over a week or two, from one or two IT people (self-reported).

Name your export owner early and the wait gets shorter. See [how long it takes to license company data](/answers/how-long-does-it-take-to-sell-company-data).

## Where does the packet go?

Counsel first. Then the free [fit check](/business) and the [buyer directory](/buyers), which lists each buyer's published terms.

Want the packet built as a dated, sourced file? [Managed readiness](/managed-readiness) does that as an optional paid service.

It turns descriptions and counts into a buyer one-pager, a systems inventory, questions for your counsel, a buyer shortlist and a term-question checklist in 2 to 3 weeks, and hands every legal call to your counsel. It can end in a no.

The [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies) covers the full process. Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Business data page](https://www.troveo.ai/business-data)
- [Data page](https://www.mercor.com/data/)
- [Handshake AI](https://joinhandshake.com/ai/data-partnerships/)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [the FTC's view](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/02/ai-other-companies-quietly-changing-your-terms-service-could-be-unfair-or-deceptive)
- [National Labor Relations Act](https://www.nlrb.gov/about-nlrb/rights-we-protect/the-law/interfering-with-employee-rights-section-7-8a1)

---

# Data monetization strategy for your company

> A data monetization strategy picks among marketplace listings, AI buyer licenses, workflows and shutdown deals. Public prices, fees and the rights check first.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/data-monetization-strategy  
Published: 2026-08-12  
Updated: 2026-10-14

A data monetization strategy is a choice among ways to earn from your records. You make it after you know what you hold and whose words are in it.

Public sources show four routes:

- list data products on an enterprise marketplace
- license operating records to AI buyers
- join a recurring workflows program
- license a closing company's archive

Your rights decide which of them are open to you.

> **Data licensing means a company gets to use your records for a while, under rules.** Think of renting a room. Each route is a different kind of guest. You keep the house either way.

This guide maps the routes and what each one publishes about price and fees. The AI-lab license itself lives in the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies). Most payout figures come from the firms selling the service.

## What does a data monetization strategy decide?

Four things: which records, which buyers, how you get paid and how much control you keep. Everything else follows.

A company that picks the buyer first can find out at the contract stage that it lacks the rights to the records that buyer wants. So work in this order. List your systems, check your rights, then pick a route.

The ten-minute sheet in [how much your business data is worth to AI](/guides/how-much-is-your-business-data-worth-to-ai) does the first step.

## Which routes exist?

| Route | Who pays | What it publishes |
|---|---|---|
| Enterprise marketplace listing | Enterprise buyers of data products | AWS lists 3% on public offers for Data Exchange products. Snowflake takes a transaction fee and does not publish the percentage |
| Corpus license | AI labs and agent builders, through a marketplace or direct | micro1's Enterprise Data Partnership listing says $100K-$2M+ for approved data packages. Handshake AI says $100K to $4M+ |
| Workflows program | micro1, per finished task | micro1 pays employees per task to grade an AI agent on real work |
| Shutdown deal | Buyers of a closing startup's archive | SimpleClosure reported nearly 100 deals at $10,000 to $100,000 |

These are self-reported or press figures. Every public deal so far is a distressed company or a buyer headline. No public record shows what a working company with a decade of history receives.

## What do marketplaces like Snowflake and AWS Data Exchange offer?

Marketplaces such as Snowflake, AWS Data Exchange, Databricks and [Datarade](/buyers/datarade) list data products for enterprise buyers. They do not buy archives.

Snowflake said that its marketplace partners booked more than $100M gross in the first half of 2026, up 277% year on year ([Snowflake](https://www.snowflake.com/en/blog/snowflake-marketplace-agentic-ai-growth/)).

Fees differ. AWS's announcement lists 3% on public offers for AWS Data Exchange products. On private offers it lists 3%, 2% and 1.5% by deal size ([AWS](https://aws.amazon.com/about-aws/whats-new/2024/01/aws-marketplace-simplified-reduced-listing-fees/)).

Datarade's commission appears as 30% in one summary and in tiers in another, so read its current terms.

The product on a marketplace is a packaged dataset with a set layout. A raw email archive is a different thing.

## What does the AI route look like?

AI data companies want records of real work. A support ticket is a good example. It shows the problem, the steps and the result. [micro1](/buyers/micro1)'s CFO described the prize in an undated recorded forum talk as the chain from request to finished deliverable.

The [buyer directory](/buyers) compares the players. [Where to license company data](/answers/where-to-sell-company-data) compares the routes.

Three payment models appear:

- a one-time payment
- a revenue share on repeat deals
- a royalty per use

[Pricing models](/answers/data-monetization-pricing) shows the public figures for each.

Plan around the gap. Epoch AI's review of RL environments describes app replicas and lab partnerships with product companies, and does not discuss real company archives ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)).

## How do you choose a route?

Score each route against five questions.

1. **Rights.** Do you hold the right to license every record in scope? Third-party confidentiality, trade secrets, privilege and employee talk that US labor law protects all limit the answer ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).
2. **Shape.** Is your data a packaged product, a pile of work records or a stream of new activity? Each fits a different route.
3. **Size.** The stated minimums differ. micro1's pages say 30 or more employees, with mature operations. [Handshake AI](/buyers/handshake-ai)'s referral page says 20 or more and 3 or more years.
4. **Control.** Who removes names, whether you may clean the data first, and what happens at the end of the term. micro1 staff said in an undated recorded forum talk that companies cannot clean first because micro1 transforms the data itself.
5. **Cadence.** One-off or recurring. In the same talk, micro1's David Revland, who works on human data, described a possible top-up about six months after a corpus deal, and a workflows track that repeats by task.

## What does it cost to carry out?

micro1's CFO estimated, in an undated talk, about 10 hours over a week or two from one or two of your IT people to run a corpus transfer. You will also need counsel. None of the buyer pages price that.

Ask each buyer who pays to remove names and who carries the blame if a record leaks. For the buyers we checked, that split is not public.

## What can go wrong?

Removing names is the weak joint. Counsel writes that stripping identifying details from millions of emails and chats is much harder than doing it for structured fields ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)). The [de-identification page](/answers/data-de-identification) covers each legal test.

The largest public deal is still contested. The Spirit Airlines sale is still before the bankruptcy court, after union objections and an EPIC brief ([EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)).

The hearing was moved from 14 Oct to 6 Nov 2026 ([Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), 9 Oct 2026). No sale order as of 10 Oct 2026. The [risk map](/risks) lists the other failure points.

Figure: steps Inventory and rights -> Pick a route -> Sign with counsel. Know what you hold before you choose who buys it.

_The last step is the one people skip. It is also the one that saves the money._

## Where to start this week

Run the inventory, then the [rights check](/rights-check), then take the free [fit check](/business). If you want help getting records ready, see the [managed readiness service](/managed-readiness).

Take your list to a privacy or commercial lawyer in your state before you sign anything. Two answers matter: a buyer's estimate and your lawyer's view on rights.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Snowflake](https://www.snowflake.com/en/blog/snowflake-marketplace-agentic-ai-growth/)
- [AWS](https://aws.amazon.com/about-aws/whats-new/2024/01/aws-marketplace-simplified-reduced-listing-fees/)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [EPIC](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/)
- [Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data)

---

# Startup shutdown checklist: where AI data fits

> A startup shutdown checklist with the AI data decision placed before you delete accounts or cancel Slack. Export clocks, approvals and buyers. Not legal advice.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/startup-shutdown-checklist-ai-data  
Published: 2026-08-12  
Updated: 2026-10-14

Decide what happens to your company data before you delete accounts, cancel Slack or let subscriptions lapse. Those steps start export clocks and delete records. A data license needs the records to still exist.

Eleven steps below run in order. The sources are vendor terms, Delaware statutes and a shutdown service's own advice. Take this checklist to your lawyer. It is not legal advice.

> **Data licensing means a company gets to use your records for a while, under rules.** Think of the attic. Before you clear the house, look in the boxes. Once the dumpster comes, they are gone.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

## What does the data decision cover?

You have four options. Keep an export for the record. License the data. Hand it to the buyer of the company in an asset sale. Or delete it. A closing company can pick more than one for different data, such as licensing chat and deleting payroll.

Some of it has buyers. [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916) reported that [SimpleClosure](/buyers/simpleclosure) had processed nearly 100 deals in a year at $10,000 to $100,000 per company.

[BigGo](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d) reported that Mercor offered up to $300,000 to failed and acquired startups. Those are company and press claims, and no outcome is promised. Our page on [AI labs buying startup archives](/answers/ai-labs-buying-startup-archives) has the reporting.

## What is the order of the checklist?

1. **Freeze deletion.** Pause account closures and auto-delete retention settings. [Slack's export help page](https://slack.com/help/articles/201658943-Export-your-workspace-data) says what you can export depends on your plan.
2. **List every system and its size.** micro1's data manifest form, as David Revland, who works on human data at micro1, described it in an undated recorded forum talk, asks which tools you use and rough sizes per tool.

   So the list doubles as your first intake conversation. See [the data manifest form](/answers/what-is-a-data-manifest-form).
3. **Find out who can export, and on which plan.** Slack's Free and Pro plans export public channels only.

   Business+ and Enterprise can request private channels and direct messages. [Zendesk exports](https://support.zendesk.com/hc/en-us/articles/4408886165402-Exporting-ticket-user-or-organization-data-from-your-account) are off by default until the account owner asks Zendesk support to enable them.
4. **Write down each export clock.** [Salesforce](https://www.salesforce.com/en-us/wp-content/uploads/sites/4/documents/legal/Salesforce_MSA.pdf) gives you 30 days after termination to ask for your Customer Data.

   Its [Data Export Service](https://trailhead.salesforce.com/content/learn/modules/lex_implementation_data_management/lex_implementation_data_export) deletes zip files 48 hours after the email goes out. Zendesk download links last at least three days. See [Salesforce](/answers/sell-salesforce-data-to-ai) and [Zendesk](/answers/sell-zendesk-tickets-to-ai).
5. **Pull the contracts.** Customer agreements, NDAs, vendor terms, investor documents and employee policies can all limit a license. Start with the [rights check](/rights-check) and our page on [NDAs](/answers/does-my-nda-let-me-sell-data-to-ai).
6. **Get the corporate approvals.** For a Delaware corporation, [section 271](https://delcode.delaware.gov/title8/c001/sc10/index.html) ties a sale of all or substantially all property to board action and a stockholder vote. Section 275 sets the dissolution vote. Read [our Delaware walkthrough](/answers/can-a-shutting-down-company-sell-its-slack-to-ai).
7. **Choose a route.** Pick a shutdown service such as the [Asset Hub](/answers/simpleclosure-asset-hub), a listing board such as [BuyMyDeadStartup](/answers/buymydeadstartup), or a direct program. Compare them in [startup data routes](/answers/sell-startup-data-to-ai) and [SimpleClosure alternatives](/answers/simpleclosure-alternatives).
8. **Check the buyer.** SimpleClosure's CEO, Dori Yona, [wrote](https://simpleclosure.com/blog/guides/how-real-are-those-million-dollar-data-offers/) that founders should look up the buyer's history. Ask whether it has completed a deal with another startup. Find out whether it uses the data or buyers the deal.
9. **Limit what you share.** The same advice says to avoid sharing full exports before meaningful offers.

   Share limited samples first, and search for embedded credentials such as API keys. It says scrubbing a dataset is more work than exporting it and deleting a few names.
10. **Get the money terms in writing.** Ask for specific prices instead of ranges, payment timing, which company is legally obligated to pay and where the funds come from. Read [reps and warranties](/answers/what-reps-and-warranties-does-an-ai-data-buyer-ask-for) and [who is liable if the buyer is hacked](/answers/who-is-liable-if-the-data-buyer-is-hacked).
11. **Record the decision, then close the accounts.** Keep the counsel notes and the board record.

    [Frankfurt Kurnit Klein and Selz](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes) says that if the company is later acquired, a prior data license will come up in diligence. An acquirer will expect to see the record. Keep it with the deal file.

_The order matters. Closing the accounts is the last box to tick, and the one you cannot untick._

## What happens to creditor claims after a startup dissolves?

A Delaware corporation continues as a legal entity for three years after dissolution to settle its business and defend suits (section 278). Section 280 sets notice to creditors, with at least 60 days to submit claims. Section 281(b) sets a plan for paying claims in priority.

Under section 281(c), directors who follow section 281 are not personally liable to claimants. We found nothing on how far that shield reaches over privacy claims.

The people in your messages are still people. Customer and partner contracts continue after the company stops. See [whether a shutting-down company can license its Slack](/answers/can-a-shutting-down-company-sell-its-slack-to-ai) for what your staff may object to.

The state and federal filings run alongside the data decision. The [state-by-state dissolution table](/closing-a-business/how-to-dissolve-an-llc-or-corporation) gives each state's form and fee, and the [final tax return when closing a business](/answers/final-tax-return-when-closing-a-business) lists the IRS forms in order.

Some records have to outlive the company. Our page on [what to do with company records when you close](/answers/what-to-do-with-company-records-when-you-close) gives the federal keep periods for tax, payroll and hiring records.

## What if the company is in bankruptcy?

A bankruptcy adds a court. In some sales it also adds a consumer privacy ombudsman. The ombudsman exists only when the debtor's privacy policy barred transferring personal information to unaffiliated parties, under [11 U.S.C. 363(b)](https://www.law.cornell.edu/uscode/text/11/363).

The Spirit Airlines sale is the live example. See [what is a consumer privacy ombudsman](/answers/what-is-a-consumer-privacy-ombudsman) and the [Spirit Airlines page](/answers/spirit-airlines-data-sale).

An insolvent company can also wind down under state law, without a bankruptcy court. Our page on [an assignment for the benefit of creditors](/answers/assignment-for-the-benefit-of-creditors) covers what happens to the data there.

## What if you do nothing with the data?

Subscriptions lapse and vendors delete it on their own schedules. The Salesforce agreement says it deletes or destroys copies after the 30-day period.

For that data the default outcome is deletion. Deleting is a valid decision too, as long as someone made it on purpose and wrote it down.

Doing nothing with the company itself costs more. Our page on [closing a business and walking away](/answers/close-my-business-and-walk-away) adds up the fees and taxes an unfiled company keeps owing.

Figure: steps Inventory the tools -> Choose a route -> Close the accounts. Pick a route for the data while the accounts are still open.

## What should you do this week?

List your systems and who can export from each. Run the [rights check](/rights-check). Then run the free [fit check](/business) to see whether buyers' published thresholds fit. If you want the file assembled for your lawyer, see our [managed readiness service](/managed-readiness).

Talk to a corporate or insolvency lawyer and a privacy lawyer in your state before you sign anything.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)
- [BigGo](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d)
- [Slack's export help page](https://slack.com/help/articles/201658943-Export-your-workspace-data)
- [Zendesk exports](https://support.zendesk.com/hc/en-us/articles/4408886165402-Exporting-ticket-user-or-organization-data-from-your-account)
- [Salesforce](https://www.salesforce.com/en-us/wp-content/uploads/sites/4/documents/legal/Salesforce_MSA.pdf)
- [Data Export Service](https://trailhead.salesforce.com/content/learn/modules/lex_implementation_data_management/lex_implementation_data_export)
- [section 271](https://delcode.delaware.gov/title8/c001/sc10/index.html)
- [wrote](https://simpleclosure.com/blog/guides/how-real-are-those-million-dollar-data-offers/)
- [Frankfurt Kurnit Klein and Selz](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [11 U.S.C. 363(b)](https://www.law.cornell.edu/uscode/text/11/363)

---

# Best AI data brokers for company records

> Best AI data brokers for company records: micro1, Troveo, Replay, SimpleClosure, Mercor, Handshake AI and more. What each asks of a company and what to check.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/best-ai-data-brokers  
Published: 2026-08-10  
Updated: 2026-10-14

No public source ranks AI data brokers, so none is best for everyone. The firms handling company records include [micro1](/buyers/micro1), [Troveo](/buyers/troveo), [Replay](/buyers/replay), [Polyshares](/buyers/polyshares), [SimpleClosure](/buyers/simpleclosure), [Mercor](/buyers/mercor) and [Handshake AI](/buyers/handshake-ai).

[Handshake AI's page](https://joinhandshake.com/ai/data-partnerships/) goes up to $4M+. None has independent proof of its payouts to companies.

This guide lists each firm, what it asks of a company and what you can check yourself. Nearly every figure comes from the firms. None is a promise of what you would receive.

> **Data licensing means a company gets to use your records for a while, under rules.** It's like renting out a room you never use. You keep the house. A broker is the agent who finds the tenant.

## What counts as an AI data broker?

In AI, people say "data broker" for any firm that sits between two sides. One side holds records, such as Slack threads, email, tickets and CRM history. The other side is the AI labs and agent builders who want them.

In law, a "data broker" resells consumer data and must register in some states. So this site calls these firms buyers, marketplaces, advisors and referral front ends.

Some take the data themselves, by purchase or by license. Some resell it on a marketplace. Some collect your details and pass them to a partner.

That choice decides who you deal with after the first form. A direct program sets the price. A marketplace finds buyers and shares the revenue. A referral front end hands you on.

Figure: comparison of Three kinds of AI data broker: Direct programs (emphasized: They are the ones who pay you, Pay you directly), Marketplaces (Resell and share), Referral front ends (Pass you on). Direct programs are the ones who pay you.

## Who handles company data, and what do they ask?

| Firm | Kind | What it asks of a company | What it publishes |
|---|---|---|---|
| [micro1](https://www.micro1.ai/data-partnerships) | Direct data partnership | 30+ employees on its pages, mostly English records, US, UK and Canada first | Tiers of $100K+, $500K+ and $1M+ |
| [Handshake AI](https://joinhandshake.com/ai/data-partnerships/) | Direct program that delivers to labs | A form, then a scoping call. [Handshake AI's company referral page](https://joinhandshake.com/ai/data-partnerships/referral) says 20+ full-time employees and 3+ years | $100K to $4M+ |
| [Mercor](https://www.mercor.com/data/) | Buyer and reseller | No minimum stated. Connects 100+ tools | Wire within about 2 to 4 weeks of signing. No price list |
| [Troveo](https://www.troveo.ai/business-data) | Marketplace | No minimum stated | Revenue share on each deal, split unpublished. No named buyers |
| [Replay](https://www.replay.ai/) | Direct program | Bands start at 20 to 50 employees | $10K to $100K for 20 to 50 staff, up to $1M+ at 250+. Upfront plus revenue share |
| [Polyshares](https://polyshares.com/) | Direct license | No minimum stated | $100K to $2M+. Net 30 to 60 after approval |
| [SimpleClosure Asset Hub](https://simpleclosure.com/product/assethub/) | Shutdown platform | Built for closing companies, open to operating ones | Nearly 100 deals at $10,000 to $100,000 (Gizmodo). A percentage on a sale or license, rate unpublished |

One more name comes up. [Protege](/buyers/protege) runs a governed marketplace in health, video and audio. Its pages do not mention workplace chat ([Protege](https://withprotege.ai/)).

_Every firm here has a friendly website. The friendly part is free._

## What can you check, and what can't you?

Some facts anyone can check.

- **People and companies:** micro1, Troveo and SimpleClosure name their leaders or companies. Troveo's footer names no legal entity.
- **Terms pages:** SimpleClosure publishes terms. A missing terms or privacy page is a reason to ask for the contract before anything else.
- **Domain age:** a lookup shows when a site was registered. A domain from this year has no track record to read.

Other facts are claims. Troveo's own pages show $20M+ and $50M+ paid. Handshake AI's show $250M and $500M+. Each firm gives two totals that its pages do not reconcile.

Troveo's $20M also covers video and audio licensors. None of these totals predicts what your company would receive. micro1's Enterprise Data Partnership listing says $100K-$2M+ for approved data packages (self-reported).

The public price points outside the companies' own published ranges are bids and distressed sales. The largest is Google's $10M winning bid for Spirit Airlines' archive, pending court approval. Mercor was at $7.5M and micro1 at $12.5M after the auction closed.

No public deal shows what a working company with ten years of CRM history receives. Our [deals tracker](/deals) follows these figures.

## Which kind of program fits which company?

Ask which one fits you. A ranking can't tell you.

- **A startup that is shutting down** fits SimpleClosure Asset Hub. It was built for that.
- **A company with 30 or more people, English-language records and years of Slack, email or CRM data** fits the direct programs that publish tiers, micro1 and Handshake AI.
- **A company of 20 to 29 people that wants a published price band** fits Replay, which starts its bands at 20 employees. Our [Replay review](/answers/replay-review) covers its terms.
- **A company that wants a marketplace to find several buyers** fits Troveo. Ask about the revenue share first.
- **A company that wants someone to prepare its records and rights first** fits a readiness service. Ours is [managed readiness](/managed-readiness).

## Who gets paid when you sign?

Some companies in this market pay people who introduce a company, and that money can shape what gets written about them. Our own fee note at the end covers every buyer we mention. Ask any comparison, this one included, who gets paid when you sign up.

## What should you do before a first call?

1. Run the [rights check](/rights-check) so you know whose words sit in your systems.
2. Read the [risk map](/risks) for failure points for data owners.
3. Work through [how to choose an AI data broker](/guides/how-to-choose-an-ai-data-broker) and [AI data broker red flags](/answers/ai-data-broker-red-flags).
4. Keep your raw data in your own systems until a signed agreement says who receives what and when it is deleted.

The [buyer directory](/buyers) records each firm's public terms. [How to license your company's data to AI companies](/guides/how-to-sell-company-data-to-ai-companies) walks through the whole process. Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Handshake AI's page](https://joinhandshake.com/ai/data-partnerships/)
- [micro1](https://www.micro1.ai/data-partnerships)
- [Handshake AI's company referral page](https://joinhandshake.com/ai/data-partnerships/referral)
- [Mercor](https://www.mercor.com/data/)
- [Troveo](https://www.troveo.ai/business-data)
- [Replay](https://www.replay.ai/)
- [Polyshares](https://polyshares.com/)
- [SimpleClosure Asset Hub](https://simpleclosure.com/product/assethub/)
- [Protege](https://withprotege.ai/)

---

# How Handshake AI uses your company data

> How Handshake AI uses your company data: what it strips out, who gets it, the license, and the re-identification warning on its own page.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-handshake-ai-uses-your-company-data  
Published: 2026-07-31  
Updated: 2026-10-14

[Handshake AI](/buyers/handshake-ai) says you hand over your company data as it exists today. Handshake says it removes identifying details.

Then the prepared dataset goes to frontier AI labs for training and testing, under contractual confidentiality. It calls the deal a license and says partner companies are never named.

> **Data licensing means a company gets to use your records for a while, under rules. The contract says who owns what.** Think of a library. AI labs borrow real work to learn how work gets done. Nobody walks out with the shelf.

Handshake has the shortest data-handling text of the buyers we read. It is also the only one of the six that admits on its own page that removing names can fail.

All of it comes from [Handshake's data partnerships page](https://joinhandshake.com/ai/data-partnerships/). The program overview is in [Handshake AI data partnerships](/answers/handshake-ai-data-partnerships).

Our only referral relationship is with micro1, an AI data company. It may pay us a fee if a company we introduce goes ahead with it. You pay nothing extra. [How we are paid](https://vaultfish.com/disclosure)

## What data does Handshake AI ask for?

[The page](https://joinhandshake.com/ai/data-partnerships/) doesn't list data types. You fill out an intake form about your company and your tools. Then you talk to a team member and hold a scoping call. The call covers data parameters, removing identifying details, terms and timeline.

The page describes the offer as operational data from businesses, with names and details removed and records formatted to spec. So the real list gets set on the call. Ask for it in writing before anything moves.

## How does Handshake AI collect it?

[The page](https://joinhandshake.com/ai/data-partnerships/) says you "transfer the data as it exists today" and that Handshake prepares it on its side. It doesn't say if that means an export, a connection or a shared drive. It doesn't say who does the work either.

It does put one duty on you in plain words. Data partners remain responsible for making sure their transfer to Handshake AI complies with applicable legal requirements.

That is the most important sentence on the page for a data owner. It puts the consent and notice question on you.

## What does Handshake AI say it does to the data?

Two statements sit side by side.

The first is a promise. [Handshake](https://joinhandshake.com/ai/data-partnerships/) applies a structured process to remove direct and indirect personal identifiers from all datasets before delivery. It takes commercially reasonable steps so the data can't be tied to a person without extra information.

The second is the admission. No removal process wipes out the risk entirely. A third party might re-identify personal information by combining it with other data. We'd weigh that line more than any promise around it.

The method, the tools and the sampling are not described. For how re-identification happens, read [can AI re-identify anonymized emails and Slack messages](/answers/can-ai-re-identify-anonymized-emails-and-slack-messages).

Figure: steps Scoping call -> Handshake says it removes names -> Labs train and evaluate. The prepared dataset reaches labs under a confidentiality contract.

## Who receives it?

Frontier AI labs, under contractual confidentiality, per [the page](https://joinhandshake.com/ai/data-partnerships/). It names no lab. It says partner companies are never named. Access is limited to authorized teams. The data is for model training and testing only, and it is not published or made public.

"Training and evaluation only" is a limit worth putting in the contract in the same words. A marketing line is not a clause.

_A friendly page is not a signed clause. Get the clause._

## Does Handshake AI license your data or take it outright?

It describes a license: "We partner with you to license your company data." The license terms are not published.

The page says nothing on exclusivity. It doesn't say if a lab can pass the data on. It doesn't say what happens to datasets built from yours. It also says nothing on retention or deletion, so those stay blank on the public record.

The page's calculator adds two notes. Its figures are an indicative estimate, not an offer to buy your data. A payout depends on due diligence and your acceptance of its licensing terms.

## What security does Handshake AI claim?

One sentence. Transfer and use are covered by your contract, with access limited to authorized teams. The page shows no certification, no encryption statement and no incident history.

We found no breach report tied to the data program. Reviews of Handshake mostly concern the worker side, not data partnerships. We cover those in [is Handshake AI legit](/answers/is-handshake-ai-legit).

## How does Handshake AI pay?

[The page](https://joinhandshake.com/ai/data-partnerships/) says partnerships "range from $100K to $4M+ depending on the data shared".

A calculator example on the page shows $383,083 to $896,969 for 100 employees and 10 years of US data, labelled indicative. All figures are Handshake's, and none promises you an outcome.

Its FAQ says payments follow defined delivery milestones. Payment typically comes 30 to 60 days after Handshake accepts the de-identified dataset.

## What should you ask Handshake AI?

- What data types does the contract cover, and can I exclude systems?
- Who removes the identifying details, with what method, and is it sampled?
- Who holds any mapping between original and replaced identifiers?
- Can the license be passed to another lab or turned into new datasets?
- What are the retention and deletion terms?
- If the dataset is re-identified, who is responsible?

The [rights check](/rights-check) covers what you need to hold before you transfer. Talk to a privacy or commercial lawyer in your state before you sign.

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Handshake's data partnerships page](https://joinhandshake.com/ai/data-partnerships/)
- [Handshake AI referral](https://joinhandshake.com/ai/data-partnerships/referral)
- [Business Insider via AOL](https://www.aol.com/articles/contractors-worked-openai-projects-handshake-050101409.html)

---

# Data anonymization before a company data license

> Searching for anonymization before a company data sale? Licensing needs it too. What the legal tests ask, why email and chat are hard, and who does the work.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/data-anonymization  
Published: 2026-07-24  
Updated: 2026-10-14

Data anonymization means changing records so no one can be picked out from them. Three facts shape the job when a company licenses an email, chat or ticket archive to an AI buyer.

1. The legal tests differ by place.
2. Free text is much harder to scrub than a database.
3. In the deals I read, the buyer or its chosen vendor does the work.

So the company's job is to ask who does it, to what standard and with what certification.

Most search results for this phrase come from software vendors. This guide is the data owner's view. It is research, not legal advice.

## What does data anonymization mean when you license an archive?

Three words get used loosely.

- **Anonymization** means a person can no longer be identified from the data.
- **Pseudonymization** swaps identifiers for labels and keeps a way back.
- **De-identification** is the American term. Its meaning depends on the statute.

Our page on [pseudonymization vs anonymization](/answers/pseudonymization-vs-anonymization) covers the first two. [What de-identification means in a data license](/answers/data-de-identification) covers the third.

> **Anonymizing records means removing what points to a person.** Think of the library. Anonymizing pulls the borrower's name card out of every book for good. Swapping names for labels keeps the card in a locked drawer.

This matters because AI buyers often want linked records. A name swapped for the same stand-in in Slack and in a spreadsheet keeps the trail from request to deliverable. That trail is pseudonymization.

_Swapping "Dana" for "Person 4" feels like privacy. Ask who holds the key._

## Which legal tests apply?

| Test | Source | The question it asks |
|---|---|---|
| Anonymous data | GDPR Recital 26 | Is the person no longer identifiable, counting all means reasonably likely to be used? |
| Deidentified | California Civ. Code 1798.140(m) | Can it be linked to a consumer, and are measures, a public commitment and recipient contracts in place? |
| Safe harbor | 45 CFR 164.514(b), health data | Are 18 listed identifier types removed, with no actual knowledge of re-identification? |
| Hashed data | FTC | Hashes can still identify users, so hashing alone is not anonymization |

GDPR Recital 26 asks what means are reasonably likely to be used. It counts cost, time and available technology ([Recital 26](https://gdpr-info.eu/recitals/no-26/)). The ICO says pseudonymised data is personal data in the hands of anyone holding the additional information. It marks its guidance as under review ([ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)).

The EDPB's draft Guidelines 02/2026 on anonymisation are open for comment until 30 Oct 2026. They address singling out, linkability and inference ([EDPB](https://www.edpb.europa.eu/public-consultations/guidelines-022026-on-anonymisation_en)).

California's test has three parts: reasonable measures, a public commitment not to re-identify, and contracts that bind every recipient ([statute text](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)). Without them, the statute treats a transfer of personal information for money as a sale. Our [CCPA page](/answers/is-selling-data-to-an-ai-lab-a-sale-under-ccpa) explains.

HIPAA's safe harbor lists 18 identifier types. It bars reliance on it where the entity has actual knowledge that the data could still identify someone ([45 CFR 164.514](https://www.law.cornell.edu/cfr/text/45/164.514)).

## Why is an email and chat archive harder than a database?

Names sit inside sentences, signatures and attachments. Context identifies people even without a name.

The law firm Frankfurt Kurnit Klein and Selz wrote that removing identifying details from millions of emails and chats is much harder than removing names from structured fields. It also wrote that removing names says whether a record links to a person. It says nothing about whether the content is confidential ([FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

Research points the same way. Staab and co-authors reported that language models inferred personal attributes from text with up to 85% top-1 accuracy. They found that text anonymization was ineffective against that inference ([arXiv 2310.07298](https://arxiv.org/abs/2310.07298)).

Lermen and co-authors reported up to 68% recall at 90% precision when matching pseudonymous accounts across platforms ([arXiv 2602.16800](https://arxiv.org/abs/2602.16800)). Both are research papers on arXiv. Neither tests a company archive. They show what a determined model can do.

The Spirit Airlines consumer privacy ombudsman wrote in a footnote that technology has "far outpaced the law" on re-identification ([Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)). Our [re-identification page](/answers/can-ai-re-identify-anonymized-emails-and-slack-messages) goes further.

## Who does the data anonymization in a deal?

In the deals I read, the receiving company, the marketplace or a vendor they accept.

**[micro1](/buyers/micro1).** Its CFO said in an undated recorded forum talk that every corpus goes through a scrubbing step for personal identifiable information, sampled at several points.

micro1's David Revland, who works on human data, said it turns names into consistent stand-ins. micro1 staff said companies cannot pre-scrub, because blanking names out would destroy the cross-platform trail.

**Spirit Airlines.** The sale agreement has the buyer accept or designate the de-identification agent ([Doc 1489](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1489.0.pdf)). The ombudsman says Tonic.ai will strip identifying details from unstructured data and file a certification before each transfer. It also says micro1 proposed to do that work itself ([Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)).

**[Handshake AI](/buyers/handshake-ai).** Its page says it removes direct and indirect identifiers before delivery. It also says a third party may still be able to re-identify personal information in combination with other data ([Handshake AI](https://joinhandshake.com/ai/data-partnerships/)).

## Should you scrub the data yourself first?

Ask before you touch anything. micro1 staff said in an undated recorded forum talk that a company's own scrub gets in its way.

FKKS advises you to stay in control. That means selecting or jointly selecting the provider, approving the protocol and requiring certification to your satisfaction.

Both can be true in one deal. The company chooses and approves the process, and the receiving company's side runs it. Settle which of those a buyer accepts before any export.

A scrub also leaves a second problem untouched. FKKS recommends leaving out privileged material, HR and personnel files, payroll, health and benefits data, union communications and customer or partner confidential information entirely.

In FKKS's view, anonymization does not make confidential content safe to license.

## What should you do before the first call?

1. Inventory systems and record types, with rough sizes.
2. Mark the categories to exclude, and read contracts and NDAs for third-party content.
3. Collect the dated privacy notices and handbooks people were given.
4. Ask each buyer who transforms the data, to what legal standard, and who certifies it.
5. Ask whether links between records survive, why, and who can reverse them.
6. Ask whether the buyer can pass the data on, and on what terms.

Our [rights check](/rights-check) covers the contract, NDA and personal-data questions in steps 2 and 3, and the [risk map](/risks) shows where deals have broken. A [managed readiness review](/managed-readiness) gathers the facts for your lawyer. Talk to a privacy or commercial lawyer in your state before you sign.

Figure: steps Exclude high-risk records -> Agree who strips names -> Certify the result. Get all three settled in writing before any data leaves.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Recital 26](https://gdpr-info.eu/recitals/no-26/)
- [ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/anonymisation/pseudonymisation/)
- [EDPB](https://www.edpb.europa.eu/public-consultations/guidelines-022026-on-anonymisation_en)
- [statute text](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)
- [45 CFR 164.514](https://www.law.cornell.edu/cfr/text/45/164.514)
- [FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [arXiv 2310.07298](https://arxiv.org/abs/2310.07298)
- [arXiv 2602.16800](https://arxiv.org/abs/2602.16800)
- [Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)
- [Doc 1489](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1489.0.pdf)
- [Handshake AI](https://joinhandshake.com/ai/data-partnerships/)

---

# How micro1 uses your company data

> How micro1 uses your company data: what it asks for, how it collects, the name swap, who sees it, ownership, deletion and pay.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/how-micro1-uses-your-company-data  
Published: 2026-07-22  
Updated: 2026-10-13

[micro1](/buyers/micro1) says it swaps the personal names in a company's workflow data, then uses the result to train and test AI agents for labs. Per [its own page](https://www.micro1.ai/data-partnerships), the company keeps ownership of the underlying data. Pay comes as a corpus deal, a recurring workflows track, or both.

> **Data licensing means a company gets to use your records for a while, under rules. You keep ownership.** Think of the attic. The boxes stay yours. micro1 gets to look through them for an agreed time.

Everything here is micro1 describing itself. It comes from its [data partnerships page](https://www.micro1.ai/data-partnerships), an undated recorded forum talk and its own video.

We could not check any of it independently, so read each line as a claim. For the program mechanics, start with [how the micro1 data partnership works](/guides/micro1-data-partnership). This guide covers only what happens to the data.

## What data does micro1 ask for?

[The page](https://www.micro1.ai/data-partnerships) says micro1 wants "operational data and business workflows." Its examples are SOPs, knowledge bases, internal documentation, CRM data, project histories and QA processes. It also lists human feedback on AI outputs in real business settings.

In the undated recorded forum talk, David Revland, who works on human data at micro1, named Google Workspace, Slack, project management tools and GitHub.

What labs want, Revland said, is the chain from a message asking for a report to the report itself. micro1's CFO said the names and vendor line items matter less than that path.

## How does micro1 collect it?

By export and a hands-on transfer. This is the sequence Revland described in the undated recorded forum talk:

1. A referral, peer intro or direct contact.
2. A first call with a data partnership manager.
3. A data manifest form listing your tools and rough sizes.
4. A payment proposal, then contract language.
5. A transfer that micro1's forward-deployed engineering team walks your IT team through.
6. micro1 verifies what arrived, then pays.

The CFO estimated, in an undated talk, about 10 hours over a week or two from one or two people in IT. A micro1 referral partner said in a micro1 video that no IT team is needed. Both are micro1 speakers, and they disagree. Ask which one applies to you.

micro1 does not say on its pages how it connects. It could be admin access, an API (a connection between two systems) or a file you hand over.

## What does micro1 say it does to the data?

[Its page](https://www.micro1.ai/data-partnerships) says personal information is removed "when relevant." Some datasets get synthetic rewrites. Those keep the workflow structure but loosen ties to the original records.

In the undated recorded forum talk, micro1's David Revland called it transformation instead of redaction. A name becomes a consistent stand-in everywhere. So a request in Slack still matches the report in Google Workspace.

micro1's CFO said every corpus is sampled at several points. micro1 staff said in the same talk that companies cannot scrub first, because blanking names would break the trail.

A micro1 partner described the same step on video as removing names, emails and phone numbers. Those are different outcomes. [The transformation page](/answers/micro1-pii-transformation) lays them side by side.

The page adds that companies get to review representative samples before prepared data is used. That is a useful promise to hold micro1 to in writing.

Figure: steps Engineers transfer -> Names swapped, sampled -> Verified, then paid. micro1 checks what arrived before any money moves.

## Who receives it?

[The page](https://www.micro1.ai/data-partnerships) says access is restricted to a limited number of authorized team members. Data runs through isolated pipelines that stop it mixing with other engagements. The page names no lab.

Press coverage of micro1's Spirit offer is the closest public account of what micro1 wants from the data.

Fortune wrote, in its own words, that an agent needs to learn how to choose actions and use tools, and that the record of those steps is a trajectory. micro1 also offered $12.5 million for that archive after the auction.

The court's privacy ombudsman noted that micro1 proposed to remove identifying details from the data itself. Details are on [the micro1 Spirit page](/answers/micro1-spirit-airlines).

## Is it a license or an outright transfer?

micro1's current page says companies keep ownership of their underlying data. Confirm licensed uses, duration, exclusivity and payment structure in the proposed agreement. In an earlier, undated talk the CFO called the corpus track a one-time sale.

In the same talk Revland mentioned a possible top-up about six months later for newer data. The referral partner in the video said the company licenses the data and keeps full access to it.

That is three descriptions of one deal. The contract decides which one is true. Ask for the grant clause before you ask about price.

## What about retention and deletion?

[The page](https://www.micro1.ai/data-partnerships) says data is kept for a limited period under agreed retention policies. On request, or at the end of an engagement, it is deleted under the agreed terms. Deletion requests are honored per the engagement. No period is given.

Deleting your files from micro1 does not reverse any training that already happened. See [deleting after training](/answers/can-i-delete-my-data-after-an-ai-company-trains-on-it).

## What security does micro1 claim?

[The page](https://www.micro1.ai/data-partnerships) says staff with access use company-managed devices and follow security controls and training requirements. It publishes no SOC 2 or ISO certificate and no encryption statement. We found no incident report.

## How does micro1 pay?

[The page](https://www.micro1.ai/data-partnerships) says pay runs from "$100k+" to "$1M+." It depends on dataset size, workflow complexity, quality, domain expertise and uniqueness. In the undated recorded forum talk the CFO said:

- Deals near $2M
- $700 to $800 per task on the workflows track

The data partnerships page tops out at "$1M+," while the [Enterprise Data Partnership listing](https://www.micro1.ai/company-referral) says $100K-$2M+ for approved data packages. A referral partner in the video said payouts arrive about 60 days after engagement. None of this is a promise to you.

_Two ceilings for one program. We'd pick the lower one for planning._

## What should you ask micro1?

- Who holds the name-to-stand-in mapping, and is it deleted?
- Does the contract say removal or swap?
- What exactly does deletion cover, and by when?
- Which labs or buyers can receive derived datasets?
- Can I see the samples before anything is used?
- Who covers a breach on your side of the pipeline?

Our [rights check](/rights-check) lists what to settle with a lawyer. [Is micro1 legit](/answers/is-micro1-legit) lays out the evidence on the company. Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [its own page](https://www.micro1.ai/data-partnerships)
- [Enterprise Data Partnership listing](https://www.micro1.ai/company-referral)
- [micro1 referral terms](https://www.micro1.ai/legal/referral-terms)

---

# What goes in a data licensing agreement?

> A data licensing agreement in plain words: the clauses law firms say to check, what buyers publish, and what to bring a lawyer. Not legal advice.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/data-licensing-agreement  
Published: 2026-07-16  
Updated: 2026-10-14

A data licensing agreement is a contract. The owner of data gives a buyer set permission to use it, usually to train or test an AI model.

> **Data licensing means a company gets to use your records for a while, under rules.** It's like renting out a spare room. The guest uses the room. You keep the house.

Seven terms do most of the work:

- what the buyer may do with the data
- which models it covers
- how long the license lasts and what outlives it
- whether it is exclusive
- what you promise about your rights
- who pays if someone else objects
- when you get paid

No buyer publishes its supplier contract. So we built this list from law-firm notes, published supplier terms and statutes. Take it to your lawyer. It is a checklist and not legal advice.

## What is a data licensing agreement?

It grants permission. On paper, it does not hand over the data forever. The buyer gets a right to use a defined set of records for a defined purpose, and you keep the rest.

The labels vary. [Troveo](/buyers/troveo) says ownership never transfers and that it licenses on your behalf ([Troveo](https://www.troveo.ai/business-data)). [micro1](/buyers/micro1) says companies keep ownership of their underlying data.

Its CFO called the corpus track a one-time sale in an earlier, undated recorded forum talk (self-reported). Confirm licensed uses, duration, exclusivity and payment structure in the proposed agreement.

[SimpleClosure](/buyers/simpleclosure)'s supplier terms grant a non-exclusive, transferable, sublicensable, perpetual and irrevocable license ([terms](https://simpleclosure.com/tos/)). A perpetual, irrevocable license looks a lot like an outright transfer. The verbs inside the contract matter more than the label on it.

For personal information, the transfer matters most. California's privacy statute defines a sale as a transfer of personal information to a third party for "monetary or other valuable consideration" ([Civ. Code 1798.140(ad)](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)).

On my reading, a revenue share is consideration under that text. Ask counsel whether your deal is a sale.

## Which clauses do published sources say to check?

Troveo's guide lists five core terms. They are permitted use, model coverage, term and survival, exclusivity, and representations and warranties. It adds indemnification (a promise to cover the other side's losses) and audit rights ([Troveo](https://www.troveo.ai/resources/ai-data-licensing-deals)).

Sidley Austin's note adds chain of title (proof you own the data) and who owns trained models. It also covers limits on passing the data on, audit and deletion ([Sidley](https://www.sidley.com/en/insights/newsupdates/2026/07/legal-implications-in-ai-development-and-deployment-training-data-and-grounding-data)).

| Clause | What it settles | Where a published example shows up |
|---|---|---|
| Permitted use | Training, fine-tuning, testing, retrieval or a mix | Troveo guide |
| Model coverage | Which models and successors, internal or commercial | Troveo guide, Sidley |
| Term and survival | How long it runs and what outlives it | Troveo guide |
| Exclusivity | One buyer or many | Troveo guide |
| Reps and warranties | What you promise about your rights | SimpleClosure terms |
| Indemnity and caps | Who pays if a third party objects | SimpleClosure terms |
| Sublicensing | Whether the buyer can pass it on | Sidley, Spirit sale |
| Audit and deletion | How you check use and what happens at the end | Sidley, CCPA |
| Payment | Amount, timing, share | Mercor, Troveo pages |

Whether the permitted use is training or retrieval changes the pricing, the term and what survives. [Training license vs RAG license](/answers/training-license-vs-rag-license) sets the two side by side.

## What happens to a trained model when the license ends?

Troveo's guide says models already trained survive the end of the license. Daniel Goldberg, a partner at Frankfurt Kurnit, says data that trains a model cannot practically be pulled back out ([FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

So a deletion clause covers your copies and the buyer's stored data. It does not undo training. Read any deletion promise with that in mind.

**Figure: Deletion covers the files, and training that already happened stays.** Two columns. Under your copies, a stack of files goes into a trash can, greyed out. Under trained model, a highlighted chip stays where it is.

## How much does exclusivity change the deal?

Troveo's guide calls exclusivity the term with the most effect on price (self-reported). It says field-of-use exclusivity is a common middle path. That means exclusive for one application only.

Epoch AI's interviews found exclusive RL-environment deals at about 4x to 5x the price of non-exclusive ones ([Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)). That is a different product from a company archive. It shows the direction. It is not your number.

Figure: comparison of Exclusivity options in a data license: Non-exclusive (Many buyers), Field-of-use (One application), Exclusive (emphasized: Biggest effect on price, One buyer, premium). Troveo's guide names exclusivity as the term that moves price most.

## What will a buyer ask you to promise in the agreement?

SimpleClosure's supplier terms are the one published example. The supplier promises that it owns the assets or has authority over them. It also promises that nothing breaks third-party or privacy rights.

The supplier also promises that it gave the required notices and got the required consents. Assets must be free of liens, and open-source duties must be disclosed. The supplier then protects SimpleClosure from lawsuits over breaches.

SimpleClosure caps its own liability at the greater of what you paid in six months or $100. Troveo's guide says the supplier's protection usually carries a cap tied to deal value. The two examples sit far apart.

For personal information, California's statute requires the transferring business to sign an agreement with the buyer. It limits purposes and requires the same level of protection. It also lets that business take reasonable steps to check how the buyer uses the data ([Civ. Code 1798.100(d)](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)).

## Who removes the names, and who signs off?

In the Spirit Airlines sale, an independent agent, Tonic.ai, strips identifying details from the unstructured data. The court's ombudsman recommends that the agent file a certification before each transfer.

The sale terms bar onward transfer unless the next party makes a similar public commitment ([Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)).

Frankfurt Kurnit's checklist says to pick the de-identification provider jointly with the company that receives the data, and to require certification to your satisfaction.

micro1 staff said in an undated recorded forum talk that micro1 transforms the data itself and that companies cannot clean it first (self-reported). Settle this clause before you sign.

## What are the payment terms?

Published payment terms vary. [Mercor](/buyers/mercor) says it wires payment when the anonymized dataset is delivered to the buyer, typically within 2 to 4 weeks of a signed agreement ([Mercor](https://www.mercor.com/data/)).

micro1's David Revland, who works on human data, said in the same talk that it pays after it checks the delivered data. Troveo says it pays a revenue share on every deal and publishes no split.

None of these is a promise to you. Put the schedule, the trigger and the share in the contract.

_A handshake deal is a great way to learn what a handshake is worth. Write it down._

## What should you bring to a lawyer?

Frankfurt Kurnit's list is a good start. Know your data by system and record type before offers arrive. Confirm your right to license the records under your customer and vendor contracts and your own software licenses.

Decide which categories to leave out for good. Examples are privileged material, HR files, payroll, health data and union communications. Write down the review.

Our [rights check](/rights-check) asks seven questions on the same ground. The [risk map](/risks) shows where objections have come from. For the full process, read [how to license your company's data to AI companies](/guides/how-to-sell-company-data-to-ai-companies).

Talk to a privacy or commercial lawyer in your state before you sign.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link. [How we are paid](https://vaultfish.com/disclosure)_

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Troveo](https://www.troveo.ai/business-data)
- [terms](https://simpleclosure.com/tos/)
- [Civ. Code 1798.140(ad)](https://cppa.ca.gov/regulations/pdf/cppa_act.pdf)
- [Troveo](https://www.troveo.ai/resources/ai-data-licensing-deals)
- [Sidley](https://www.sidley.com/en/insights/newsupdates/2026/07/legal-implications-in-ai-development-and-deployment-training-data-and-grounding-data)
- [FKKS](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [Epoch](https://epoch.ai/gradient-updates/state-of-rl-envs)
- [Doc 1684](https://storage.courtlistener.com/recap/gov.uscourts.nysb.331039/gov.uscourts.nysb.331039.1684.0.pdf)
- [Mercor](https://www.mercor.com/data/)

---

# Can your business license its content to AI?

> For businesses with courses, a member community, a podcast or a video archive: who licenses video and audio, and the consent to settle first.

By Matt Burns, Vaultfish. Canonical: https://vaultfish.com/guides/license-your-business-content-to-ai  
Published: 2026-07-08  
Updated: 2026-10-14

Video has the clearest buyers for business content so far. Broadcasters, producers and distributors license footage through marketplaces such as Troveo and Protege, and Bloomberg reported buyers pay about $1 to $4 a minute.

Before you license anything, check who appears in the footage and who owns the words. Community members, coaching clients and podcast guests rarely agreed to AI use.

> **Data licensing means a company gets to use your content for a while, under rules.** Think of renting out a room. The buyer stays for a term and follows your house rules. You keep the house.

Licensing your business records instead? Start with the [pillar guide on licensing company data to AI companies](/guides/how-to-sell-company-data-to-ai-companies). This page covers content your business owns: video, audio, posts, courses and sessions.

micro1 may pay us a referral fee if a company we introduce goes ahead with it. [How we are paid](https://vaultfish.com/disclosure)

## Which programs take business content for AI training?

**Video.** [Troveo](https://www.troveo.ai/content-owners) pays content owners a revenue share on gross earnings. It has not published the percentage ([Publish Press](https://news.thepublishpress.com/p/how-to-make-money-when-ai-uses-your-content)). Its content-owners page prefers 1080p or 4K and says processing takes 1 to 2 weeks. [Troveo](/buyers/troveo) also runs a separate business-data program.

[Protege](/buyers/protege) works with broadcasters, producers and distributors. They license content they own outright or control with broad rights. No minimum is stated and the share is undisclosed ([Protege](https://withprotege.ai/data-provider/video)).

Calliope Networks said it needed at least 25,000 hours of content ([Tubefilter](https://www.tubefilter.com/2024/10/01/calliope-networks-generative-ai-scraping-license-startup/)). Protege bought it on 18 Dec 2024 ([Protege](https://withprotege.substack.com/p/protege-acquires-calliope-networks-bc7)).

**New footage of hands-on work.** [micro1's video data program](/answers/micro1-commercial-video-data-partner) asks companies with at least 20 employees to film first-person video of physical work. That is new footage recorded for micro1, so your staff are the people on camera.

**Audio.** There is no public per-hour rate yet. AIPodcast.io offers back-catalog licensing with a quarterly revenue share and written guest re-clearance. It publishes no rates ([AIPodcast](https://www.aipodcast.io/solutions/podcast-audio-licensing)).

**Text.** Pay-per-use is the live model, and it favors sites with traffic. ProRata says it pays publishers 50% of all revenue from its Gist.ai product ([News/Media Alliance](https://www.newsmediaalliance.org/prorata-licensing-partnership/), [WAN-IFRA](https://wan-ifra.org/2025/01/prorata-aims-to-be-pro-publisher-when-it-comes-to-revenue-sharing-on-ai-platforms/)).

**Courses, community content and coaching transcripts.** No named buyer pays for these in public reporting so far.

## What do buyers pay for video?

AI companies pay about $1 to $4 per minute of video. For ordinary YouTube-style footage it is $1 to $2 ([Bloomberg via The Star](https://www.thestar.com.my/tech/tech-news/2025/01/13/youtubers-are-selling-their-unused-video-footage-to-ai-companies)). Those are buyer prices.

Public sources don't show how much of that price reaches the business that owns the footage. Troveo's revenue share is unpublished, and so is Protege's.

_The buyer price is the headline. Your share of it is the number to get in writing._

## Who owns the words and faces in your content?

Platform terms say what the platform may do with your content. They don't decide what you may license to a third party. That depends on your copyright and on releases from everyone in the content.

**Figure: The footage is yours to license once everyone in it has signed.** A video frame with three people in it, labelled member, client and guest. Under each person is an empty highlighted box and a signature line, labelled releases.

| Platform | What the terms say |
|---|---|
| [Skool](https://www.skool.com/legal?t=terms) | Content is yours. Skool gets a sublicensable license to provide, develop and improve its services. No AI clause, and the terms don't address admin rights over members' posts |
| [Circle](https://www.circle.so/terms) | Irrevocable, sublicensable license to process your content and use it, in non-identifying form, to develop and improve Circle's products |
| [Kajabi](https://kajabi.com/policies/terms) | You keep your content and stay responsible for the consents your students' personal data needs. You may not use MCP data to train AI without Kajabi's written consent |
| [Substack](https://substack.com/pa) | Your license to Substack does not permit it to train third-party AI models |
| YouTube | Google trains Gemini and Veo on YouTube content and channel owners cannot opt out ([CNBC](https://www.cnbc.com/2025/06/19/google-youtube-ai-training-veo-3.html)) |

The Substack clause limits Substack. As we read it, it says nothing about what you may license from your own posts.

**Community members.** Members own their posts. As we read the terms, a business that licenses its Skool or Circle archive is licensing something the terms do not hand over. There's no public case of this yet, so the risk is untested.

**Coaching clients.** Sessions carry confidentiality duties. Tandem Coach reads the ICF's Standard 2.5 on boundaries as meaning clients should know when AI tools touch the coaching relationship.

It suggests disclosure in writing and consent spoken at intake. The standard itself doesn't mention AI ([Tandem Coach](https://tandemcoach.co/ai-disclosure-coaching-clients/)).

No public market for these transcripts exists yet. Removing names from conversational text is harder than stripping fields ([Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)).

**Podcast guests.** Most signed releases before AI licensing existed. AIPodcast re-clears guests in writing and drops episodes where a guest declines or cannot be reached.

**People on camera.** Content with identifiable people needs signed waivers ([AI Optimist](https://www.theaioptimist.com/p/the-5m-ai-video-licensing-wave-why)). That includes staff who appear in training videos or webinars.

## What should you do before licensing anything?

1. **Inventory what the business owns.** List its video, audio, courses and posts by type, and who made each: staff, contractors or guests.
2. **List the people.** Guests, clients, members and anyone on camera or named. This list decides what is licensable before a buyer enters the picture.
3. **Collect what each person signed.** Releases, guest agreements, community terms, client contracts and contractor agreements. Where nothing mentions AI, treat it as an open question for a lawyer.
4. **Flip the switches you control.** YouTube third-party training is off unless you turn it on. Substack's "Tell AI tools not to train" toggle applies only to the AI tools that respect it ([Substack Support](https://support.substack.com/hc/en-us/articles/20382615953556-How-can-I-block-AI-from-using-my-Substack-publication-to-train-their-models)). Spawning runs a Do Not Train registry ([Spawning](https://spawning.substack.com/p/the-spawning-guide-to-rights-reservations)).
5. **Ask the contract questions.** Revenue share, exclusivity, term, revocation, and whether your files pool into dataset-wide deals you never see. The [pillar guide](/guides/how-to-sell-company-data-to-ai-companies) has the full checklist.
6. **Talk to a lawyer and an accountant.** Pick a lawyer who handles media or IP licensing, and see them before you sign. An accountant can explain how any payout is taxed. We are neither.

## How do you do this safely?

Run the [rights check](/rights-check) before you list anything with a buyer. It asks seven questions about whose words and faces sit in your content and what each person signed.

The [risk map](/risks) covers the legal and security failure points for data owners. The [buyer directory](/buyers) and [deals tracker](/deals) show who pays what, with dates. Keep your files in your own storage until a signed agreement says who receives them and when they are deleted.

Your business may also hold years of email, CRM or ticket records. The [readiness check for business data](/guides/how-much-is-your-business-data-worth-to-ai) takes ten minutes. The free [fit check](/business) comes next.

Do step two above first, today. The list of people in your content sets the limit on everything else.

[Apply to micro1's data partnership (paid link)](https://req.micro1.ai/post/74520e8b-4673-4da5-b855-53609aaf6710?referralCode=f82e519e-5d64-4e0e-9574-760c84371cf2&utm_source=referral&utm_medium=share&utm_campaign=data_referral) · [micro1's own data-partnerships page](https://www.micro1.ai/data-partnerships)

_Paid link: micro1 may pay us a referral fee if a company we introduce goes ahead with it. You pay nothing extra._

_Vaultfish helps you run the process. We are not lawyers and this is not legal advice. If you need counsel who knows this area, we can point you to one._

## Sources

- [Troveo](https://www.troveo.ai/content-owners)
- [Publish Press](https://news.thepublishpress.com/p/how-to-make-money-when-ai-uses-your-content)
- [Protege](https://withprotege.ai/data-provider/video)
- [Tubefilter](https://www.tubefilter.com/2024/10/01/calliope-networks-generative-ai-scraping-license-startup/)
- [Protege](https://withprotege.substack.com/p/protege-acquires-calliope-networks-bc7)
- [AIPodcast](https://www.aipodcast.io/solutions/podcast-audio-licensing)
- [News/Media Alliance](https://www.newsmediaalliance.org/prorata-licensing-partnership/)
- [WAN-IFRA](https://wan-ifra.org/2025/01/prorata-aims-to-be-pro-publisher-when-it-comes-to-revenue-sharing-on-ai-platforms/)
- [Bloomberg via The Star](https://www.thestar.com.my/tech/tech-news/2025/01/13/youtubers-are-selling-their-unused-video-footage-to-ai-companies)
- [Skool](https://www.skool.com/legal?t=terms)
- [Circle](https://www.circle.so/terms)
- [Kajabi](https://kajabi.com/policies/terms)
- [Substack](https://substack.com/pa)
- [CNBC](https://www.cnbc.com/2025/06/19/google-youtube-ai-training-veo-3.html)
- [Tandem Coach](https://tandemcoach.co/ai-disclosure-coaching-clients/)
- [Frankfurt Kurnit](https://technologylaw.fkks.com/post/102o455/selling-company-data-for-ai-training-what-to-know-before-you-say-yes)
- [AI Optimist](https://www.theaioptimist.com/p/the-5m-ai-video-licensing-wave-why)
- [Substack Support](https://support.substack.com/hc/en-us/articles/20382615953556-How-can-I-block-AI-from-using-my-Substack-publication-to-train-their-models)
- [Spawning](https://spawning.substack.com/p/the-spawning-guide-to-rights-reservations)

---

# Publicly documented company data deals and price claims

By Matt Burns, Vaultfish. Published 9 Oct 2026. Updated 16 Oct 2026. Data through 10 Oct 2026. Page: https://vaultfish.com/research/what-company-data-sells-for

## The short answer

Public figures are few, and they are different kinds of evidence, so we keep them in separate groups. The one named completed sale, Cielo24, reportedly went for "hundreds of thousands of dollars". The largest figure a court has seen is Google's $10M winning bid for Spirit Airlines' archive. No sale order as of 10 Oct 2026. Most other figures are ranges companies publish about their own programs.

## Five findings

1. **One named completed sale.** Cielo24, a defunct company, reportedly sold 13 years of Slack, email and Jira history for "hundreds of thousands of dollars". The buyer is unnamed and the report is secondary ([Forbes](https://www.forbes.com/sites/annatong/2026/04/16/ais-new-training-data-your-old-work-slacks-and-emails/)).
2. **The biggest court figure is still a bid.** Google's $10M bid won the 14 Aug 2026 auction for Spirit Airlines' archive: about 100 million emails, 500 million Teams items and 30 million lines of code ([court docket](https://www.courtlistener.com/docket/71233134/spirit-aviation-holdings-inc/), [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/)). The hearing was moved from 14 Oct to 6 Nov 2026 ([Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), 9 Oct 2026). No sale order as of 10 Oct 2026.
3. **Three offers, one archive.** Mercor was the $7.5M alternate bidder at that auction. micro1 made a late $12.5M offer after it closed ([Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/), 14 Sep 2026). We mark the three Spirit rows as related observations of one archive. Count them as one case.
4. **Published ranges are marketing claims.** SimpleClosure says it has done nearly 100 archive deals in about a year at $10,000 to $100,000 per company, by its own count ([Gizmodo](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916)). Handshake AI's ceiling is "$4M+", which it labels an indicative estimate and not an offer ([Handshake AI](https://joinhandshake.com/ai/data-partnerships/)). No named company has confirmed a figure in this group.
5. **Two rows rest on a court record.** Both are Spirit bids. Every other row rests on a press report or on the company's own figures ([court docket](https://www.courtlistener.com/docket/71233134/spirit-aviation-holdings-inc/)).

Figure: scale, Spirit Airlines archive auction, 14 Aug 2026: Mercor, runner-up $7.5M, Google, winning bid $10M (emphasized). Google outbid Mercor for Spirit Airlines' archive. Source: Time and Fortune, from the court docket.

## What these examples cannot establish

- They are not a price list. A handful of public figures, mostly bids and published claims, cannot set a market rate.
- They do not estimate what any company would be paid. Your records, rights and terms will differ from every row here, and the guide to [how much company data is worth](https://vaultfish.com/guides/how-much-is-your-business-data-worth-to-ai) covers what moves the figure.
- Published ranges are self-reported marketing claims. Treat them as a company's pitch until a named company that took part confirms a figure.

## Method

Every row is a row in the deals tracker (https://vaultfish.com/deals), frozen on the date above.

**Where we searched.** Court dockets through the CourtListener RECAP archive (the Spirit Airlines case, No. 25-11897), national, business and trade press, and the pages AI data companies and marketplaces publish about their own programs.

**What gets a row.** A court filing or news report that records a price, bid or offer for company or user data sold or licensed for AI training, a price range an AI data company or marketplace publishes about itself (marked self-reported), or a named seller's sale plan that drew a lawsuit, even with no price (Photobucket).

**What stays out.** Whole-company acquisitions (they price the business, not the data), publisher and media licensing deals, private figures told to us, and any number we cannot trace to a filing or a report. Reddit's reported $60M a year license with Google stays in, in the context group, because it is the best-known price anchor. It is a content license and neither company confirmed the figure.

**How we group rows.** Each row sits in one group by its status: completed sales, bids and offers (none closed), published price ranges, or context. Rows that are offers for the same archive are marked related, so three Spirit bids read as one case.

**Prices stay as the source states them.** We never average a range, turn a phrase like "six figures" into a number, or work out a price per employee. The study has no totals, averages or medians, inside a group or across groups. Dates are event dates. micro1's tiers page carries no date, so its row uses the day we read it. Events run from 22 Feb 2024 to 6 Oct 2026.

**Known gaps.** Private deals under confidentiality terms never reach the public record, so this set undercounts them. Two rows rest partly on reports we could not open (Forbes and Fast Company for Cielo24, Reuters for Photobucket), and those rows say so in their notes. We read the Spirit docket through the RECAP mirror, which lags the court. The 9 Oct 2026 notice that moved the Spirit hearing was not on it when we looked, so that row cites Travel Weekly (https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data) for the move.

## The dataset, by group

11 rows in four groups, oldest first inside each. CSV with a category column, notes and every source: https://vaultfish.com/research/what-company-data-sells-for.csv

### Completed sales (1 row)

A press report, or the buyer itself, says a deal closed. No court or party has published the contract.

| Date | Buyer or program | Records from | Figure as stated | Status | Evidence | Records covered | Related | Sources |
|---|---|---|---|---|---|---|---|---|
| 16 Apr 2026 | Undisclosed AI buyer | Cielo24 (defunct, via SimpleClosure) | "Hundreds of thousands of dollars", as reported | Reported completed | Secondary source only | 13 years of Slack messages, Email, Jira tickets, Multi-terabyte Google Drive archive | None | [Forbes (we could not open it, details via search summaries)](https://www.forbes.com/sites/annatong/2026/04/16/ais-new-training-data-your-old-work-slacks-and-emails/), [Fast Company (we could not open it)](https://www.fastcompany.com/91528808/shuttered-startups-are-selling-old-slack-chats-and-emails-to-ai-companies), [Strategic Signal deal tracker entry](https://strategicsignal.ai/ai-data-transactions/deals/ss-aidt-2026-0001/) |

### Bids and offers, none closed (4 rows)

Figures offered for data. None of these has closed. Rows that share an archive are related observations of one archive, not independent deals.

| Date | Buyer or program | Records from | Figure as stated | Status | Evidence | Records covered | Related | Sources |
|---|---|---|---|---|---|---|---|---|
| 14 Aug 2026 | Google | Spirit Aviation Holdings (Spirit Airlines, in Chapter 11) | $10M winning bid | Pending court approval | Court record plus press | About 100 million emails, About 500 million Teams items, About 30 million lines of code, Employee and payroll records | One of 3 related observations of the Spirit Airlines archive, not an independent deal. | [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/), [Bloomberg Law: ombudsman report](https://news.bloomberglaw.com/ip-law/spirit-data-sale-to-google-passes-muster-privacy-ombudsman-says), [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/), [EPIC amicus brief announcement](https://epic.org/epic-files-amicus-brief-challenging-spirit-airlines-bankruptcy-sale-of-employee-data-to-google/), [Travel Weekly: hearing moved to 6 Nov 2026](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), [Court docket, In re Spirit Aviation Holdings, No. 25-11897 (Bankr. S.D.N.Y.)](https://www.courtlistener.com/docket/71233134/spirit-aviation-holdings-inc/) |
| 14 Aug 2026 | Mercor | Spirit Aviation Holdings (Spirit Airlines, in Chapter 11) | $7.5M alternate bid | Bid, not accepted | Court record plus press | Same Spirit archive as the Google bid | One of 3 related observations of the Spirit Airlines archive, not an independent deal. | [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/), [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/) |
| 17 Aug 2026 | Mercor | Shuttered and acquired startups (for example Warmly, per the report) | Offers of up to $300,000 reported | Offers reported | Press report | Slack logs, GitHub records, Asana files, Google Drive documents, Meeting transcripts and recordings | None | [BigGo editorial](https://finance.biggo.com/news/d34746c0-e437-4855-93c5-09c6e850550d) |
| 3 Sep 2026 | micro1 | Spirit Aviation Holdings (Spirit Airlines, in Chapter 11) | $12.5M reported offer | Bid, not accepted | Press report | Same Spirit archive as the Google bid | One of 3 related observations of the Spirit Airlines archive, not an independent deal. | [Fortune](https://fortune.com/2026/09/14/micro1-tries-to-beat-googles-bid-to-buy-bankrupt-spirit-airlines-data-for-12-5-million-highlighting-value-of-data-to-ai-companies/), [Time](https://time.com/article/2026/08/25/google-spirit-airlines-ai-data-RL/) |

### Published price ranges (4 rows)

Ranges that AI data companies and marketplaces publish about their own programs. They are self-reported marketing claims, and no named company has confirmed a figure.

| Date | Buyer or program | Records from | Figure as stated | Status | Evidence | Records covered | Related | Sources |
|---|---|---|---|---|---|---|---|---|
| 17 Apr 2026 | Undisclosed AI companies and labs | Shut-down startups, through SimpleClosure Asset Hub | $10,000 to $100,000 per company, nearly 100 deals in about a year | Self-reported range | Self-reported by the company | Slack, email and workspace data, Source code | None | [Gizmodo, citing Forbes](https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916), [SimpleClosure Asset Hub announcement](https://simpleclosure.com/blog/insights/simpleclosure-asset-hub/), [SimpleClosure CEO post](https://simpleclosure.com/blog/guides/how-real-are-those-million-dollar-data-offers/) |
| 11 Sep 2026 | Troveo's buyers (unnamed) | Companies licensing through Troveo | Typical full-company deals start at six figures. Example dataset: $850,000 across four sales. Single repository: about $5,000, an unnamed platform's baseline it cites. Startup archives: $10,000 to $100,000 | Self-reported range | Self-reported by the company | Business data from about 30 systems, Source code | None | [Troveo: how much is company data worth](https://www.troveo.ai/resources/how-much-is-company-data-worth), [Troveo business data](https://www.troveo.ai/business-data) |
| 2 Oct 2026 | Handshake AI | Not stated | $100K to $4M+. Calculator example: $383,083 to $896,969 for 100 employees and 10 years | Self-reported range | Self-reported by the company | Operational data from businesses (types not itemized) | None | [Handshake AI data partnerships page](https://joinhandshake.com/ai/data-partnerships/), [Handshake AI data partnerships criteria page](https://joinhandshake.com/ai/data-partnerships/referral) |
| Undated | micro1 | Not stated | Tiers of $100K+, $500K+ and $1M+. Enterprise Data Partnership listing: $100K-$2M+ for approved data packages | Self-reported range | Self-reported by the company | Slack, Email, CRM, Tickets, SOPs and knowledge bases | None | [micro1 data partnerships page](https://www.micro1.ai/data-partnerships), [micro1 Enterprise Data Partnership listing](https://www.micro1.ai/company-referral) |

### Context (2 rows)

Announced plans and unconfirmed reports. They show how the market behaves, and none is a confirmed company data sale.

| Date | Buyer or program | Records from | Figure as stated | Status | Evidence | Records covered | Related | Sources |
|---|---|---|---|---|---|---|---|---|
| 22 Feb 2024 | Google | Reddit | About $60M a year, reported | Reported, unconfirmed | Press report | User-generated posts and comments | None | [The Decoder, citing Reuters](https://the-decoder.com/reddit-signs-60-million-annual-training-data-deal-with-google/) |
| 8 Apr 2024 | Unnamed technology companies (talks reported) | Photobucket | Not disclosed in the sources we could open. Press coverage of the CEO's remarks cites 5 cents to $1 per photo, unverified | Announced, no price | Secondary source only | About 13 billion user photos and videos | None | [Class Action.org on Pierce v. Photobucket, filed 11 Dec 2024](https://www.classaction.org/news/photobucket-facing-lawsuit-over-plan-to-sell-13-billion-user-photos-to-third-parties) |

## Corrections

Newest first.

**11 Oct 2026**

- Retitled to "Publicly documented company data deals and price claims" (it was "Publicly documented company data deals and buyer price claims"). The group of self-published ranges is now "Published price ranges" (it was "Buyer marketing ranges") and the bids group is now "Bids and offers, none closed". The ranges group includes micro1's partnership tiers, and micro1 runs data partnerships in which companies keep ownership, so the old label described it wrongly.
- Renamed four CSV columns to match the table headings: "buyer" is now "buyer_or_program", "data_sold" is now "records_covered", "seller" is now "records_from" and "price_as_stated" is now "figure_as_stated". Some rows are programs, such as micro1's company-data partnerships, where the company keeps ownership of its data.

**10 Oct 2026**

- Retitled to "Publicly documented company data deals and buyer price claims". The old title claimed complete coverage, which a search like ours cannot promise. The address stays the same.
- Split the rows into separate groups: completed sales, bids and pending court sales, buyer marketing ranges, and context. The CSV gains category and related_to columns.
- The Spirit Airlines hearing was moved from 14 Oct to 6 Nov 2026 ([Travel Weekly](https://www.travelweekly.com/Travel-News/Airline-News/Court-delays-hearing-on-Google-purchase-of-Spirit-data), 9 Oct 2026). No sale order as of 10 Oct 2026.
- Removed the price-per-lost-job comparison for the Spirit bids from our guides and answers. This study never printed it. Also removed a chart here that set Google's bid against SimpleClosure's top self-reported price.
- Updated the micro1 wording to describe company-data partnerships in which companies keep ownership of their data.

## How to cite this study

Burns, M. (2026). Publicly documented company data deals and price claims. Vaultfish. Published 9 Oct 2026, data through 10 Oct 2026. https://vaultfish.com/research/what-company-data-sells-for

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Reuse freely and credit Vaultfish with a link to the study.

## For journalists

Matt Burns can walk you through any row and give a comment on the record. Press inquiries: https://vaultfish.com/contact?topic=press