Company Data Partnership Programs: micro1, Mercor, Handshake and the Rest

A company data partnership program is a standing offer by an AI data company to license a business's internal records, such as Slack threads, email, tickets, code and documents, de-identify them and pass them on for AI training. The directory below covers 41 companies, 28 of which have a program that takes in company records, code, medical records or media, and three of which, micro1, Mercor and Handshake AI, publish detailed terms.

Each program sets its own eligibility, payment and referral rules, and every figure here is the program's own claim, checked 2026-11-09. This page is part of the independent reference on AI training data licensing: the order is alphabetical, listing never depends on referral fees, and DataPacts is not a broker, takes no cut of any sale and has joined no referral program yet.

Each number carries its source and check date, and how we source, date and check every figure is set out in the methodology.

What a Company Data Partnership Program Is

A program licenses de-identified operating records from many companies on one published process and passes them to AI labs, which the program pages rarely name. Mercor names Slack, Gmail, Salesforce, GitHub, Notion, Confluence, Drive, Outlook, MS Teams and "100+ supported tools", while micro1 states it takes company operating data and agent trajectories, checked 2026-11-09.

A program differs from a one-off deal or a marketplace listing in four ways: the offer is standing, a program page publishes the process, an intake form starts it, and a payout claim is stated before any negotiation. micro1 states a commitment of USD 1 billion to enterprise data acquisition over 12 months, announced 2026-10-09, which is a commitment and not money already paid.

In the buyer chain, programs are one answer to who buys AI training data from companies. micro1 states that it licenses the data to AI labs and enterprises in the US and allied countries, checked 2026-11-09.

Every Company Data Partnership Program on Record

The directory lists all 41 companies profiled under this category, alphabetically, with the program each one runs and what that program takes in. The table is as of 2026-11-09, and the Checked column carries that date wherever a program was found.

• "No record" means no company data program was found on the check date, not proof that none exists.
• "Identity not confirmed" means the company is named but nothing about it is stated here.
• "Placement" and "Referral only" rows handle data or refer holders, they do not acquire it.

Scope counts: Company records 22, Code and games 2, Medical records 1, Media and datasets 3, Placement 1, Referral only 1, No record 9, Identity not confirmed 2.

Who Runs These Programs: Human-Data Companies, RL-Environment Builders, Vendors and Intermediaries

Most programs are run by human-data companies that already sell expert work to AI labs and added company records as a second product. The same human data companies that pay expert contributors appear here, and the 41 companies fall into five organization types.

• Human-data platforms: 14
• Data vendors: 10
• Company identity not fully confirmed: 8
• RL-environment builders: 5
• Marketplaces: 4

Intermediaries sit beside the buyers. Endless names micro1, Mercor and Handshake as the buyers it places data with, states a fixed payment agreed before anything is shared and no commission on the seller's side, and SourceX publishes only a referral partner program. Sureform states that it identifies buyers and manages licensing for the company. None of the three is a buyer in this directory's sense.

Some builders turn company records into training worlds, a market covered with RL environments and who buys them.

The Three Programs in the Title: micro1, Mercor and Handshake AI, Dated

micro1, Mercor and Handshake AI publish the most detailed company data offers in the directory, and each publishes them differently: micro1 its license terms, Mercor per-industry listings, Handshake a headline range. The table sets the three programs side by side as of 2026-11-09, every value the program's claim.

Each program contradicts itself somewhere. Handshake states "Earn $100K to $4M+" in its headline and "up to $100,000" in its estimator on the same page, checked 2026-11-09. micro1 gives two payment timings on one page, and Mercor gives two as well. The contradictions stay visible as two claims, because neither can be resolved from public sources.

The summary leaves out each program's exclusivity, deletion and retraining terms and the other programs' figures, which program terms compared sets out with dates for every program.

How the Programs Work: Application, Data Review, Payment and Referrals

Nearly every program follows the same five stages: intake, scoping under an NDA, de-identification and transfer, review and acceptance, then payment, with a referral track beside it. The table lists each stage with examples as programs state them, checked 2026-11-09.

Application and Eligibility

Programs state headcount and history thresholds that differ by program and sometimes by page of the same program. micro1 states 30+ full-time employees and data primarily in English, with the US first and then other Western markets. Handshake AI states 20+ full-time employees and 3+ years on its license and referral pages and no minimum on its Data Partnerships page. Mode states 30+ on its home page, 50+ on its referral page and 50 office-based in its agreement. Nyne states 20+ US employees and 3+ years, Scale AI prefers 40+, Collinear AI works with teams of about 50+, and SuperAnnotate states "Any size, any industry, anywhere", all checked 2026-11-09.

Records of a shut-down or defunct company qualify at several programs. Turing accepts active, winding-down and wound-down companies, AfterQuery welcomes operating, winding-down and closed ones, Labelbox states distressed and wound-down companies are eligible, and Ooak Data states wind-down deals are signed by a director, liquidator or administrator.

Data Review, De-identification and Acceptance

Most programs take on de-identification themselves, and payment usually waits until the program has reviewed and accepted the delivered data. micro1 states companies need not anonymize before transfer, Mercor states that engineers do not manually review the data, Turing covers de-identification costs, and Polyshares names a specialist PII partner.

Acceptance rules differ, and sometimes within one program. micro1's Data License Master Terms (updated 2026-11-08) give 30 business days to reject, after which the data is deemed accepted (s.1.3), while its program page says "30 days". Patronus AI states it pays on acceptance and Turing that it reviews within 5 business days.

Review cannot fix everything. AfterQuery runs a re-identification test before delivery and states that "no process removes all risk on its own"; what PII removal and de-identification can and cannot achieve is covered on its own page.

How Programs Structure Payment

Programs pay in one of five ways, and the program page usually says which: a fixed license price, an upfront fee plus a revenue share, a revenue share alone, a payment per relicense, or a payment in two parts. The structures below are as the programs state them, checked 2026-11-09.

• Fixed license price: Endless, micro1
• Upfront fee plus revenue share: Labelbox, Origin Lab, Replay
• Revenue share alone: Protege, Turing
• Payment per relicense: SuperAnnotate
• Payment in two parts: Abaka AI

Milestones and clawbacks sit with when and how AI data buyers pay.

Referral Models: When a Referral Fee Is Earned

Most programs pay a fee to whoever introduces a company, but the fee is earned at different moments, and several programs pay nothing until money has moved on the deal. The triggers below are examples as programs state them, checked 2026-11-09.

• Company signs: Scale AI
• Deal closes, the purchase completes or the record sells: Handshake AI, HUMXN, Specific Labs, Turing
• Data approved, after registration through the referrer's link: micro1
• Company paid first: Patronus AI, Replay
• Program paid by its own customer first: Mode
• No referral published: Endless, Labelbox, Ooak Data, SuperAnnotate

Fee amounts sit with the referral programs of AI data buyers.

Benefits, Risks, Downsides and Alternatives of Joining a Program, Including Not Licensing

A program offers a published process and, in most cases, de-identification at the program's cost, but the license it takes is usually broad and the payout figures are claims, not confirmed payments. The lists below use the programs' own statements, checked 2026-11-09.

Benefits

• A published process: Mercor states it covers extraction, anonymization and transfer costs, with no setup fees or deductions.
• De-identification by the program: micro1 states it de-identifies sensitive information itself.
• No upfront cost: Turing states a revenue share with nothing upfront.

Risks

• Broad license: micro1's Data License Master Terms grant a nonexclusive, irrevocable, perpetual, sublicensable license (s.1.2) and set no duty to delete or retrain Licensee Materials (s.4.4).
• Exclusivity: Labelbox states Prism is exclusive by default and that trained models "cannot be unlearned".
• Security: Mercor reported unauthorized access through compromised LiteLLM versions on 2026-03-24 and 2026-03-30, and notified the California Attorney General on 2026-06-25. The incident is tracked with other data buyer incidents.
• Who de-identifies: in the Spirit Airlines case (Bankr. S.D.N.Y. No. 25-11897 (SHL)), micro1 filed a notice of intent to bid (Dkt. 1556, 2026-09-03) and proposed to de-identify the data itself, and the Consumer Privacy Ombudsman asked to review this (Dkt. 1684, 2026-10-05).

Downsides

• Unconfirmed amounts: no company has publicly confirmed an amount received under any program.
• Mixed signals: Handshake AI shows a headline and an estimator maximum that differ, and Nyne labels its estimator "indicative and not an offer".
• Slow and conditional: Mode states about three months from first conversation to payment and a 12-month clawback window on referrals.

Alternatives

• A one-off license negotiated with a buyer directly.
• A broker or marketplace that places the data.
• A sale in insolvency: on 2026-08-14 the Spirit Airlines auction of de-identified data named Google LLC the successful bidder at USD 10,000,000 and Mercor.io Corporation the alternate bidder at USD 7,500,000 (Dkt. 1463). The sale is not approved as of 2026-11-09, with a hearing set for 2026-11-17 (Dkt. 1779).
• Not licensing at all, which remains a valid outcome.

Before signing with any program, read the AI data licensing agreement clause by clause, because program pages rarely state license scope, term, exclusivity or deletion.

Where Programs Sit Among Buyers, Terms, Deals and the Seller Path

Programs are one route among several by which data reaches AI developers; the pages below cover the others and the steps around them. DataPacts explains how DataPacts is funded and why no program pays for its place.

• Tools for AI data sellers compare buyers and estimate terms.
• The full path to sell or license company data to AI companies runs from readiness to payment.
• AI data licensing deals lists one-off deals with dates and sources.
• Administrators can sell data to AI buyers in insolvency, where the Spirit auction is unapproved as of 2026-11-09.
• Head to head: micro1 vs Mercor, Mercor vs Handshake AI and micro1 vs Handshake AI for company data.
• Beyond programs: alternatives to selling data, choosing a buyer for your data and reported payouts to data sellers.

Questions About Company Data Partnership Programs

Which company data partnership program is best?

No program is best for every company; fit depends on the records held, headcount, country and the license terms the company will accept. DataPacts does not rank programs, and thresholds differ: micro1 states 30+ full-time employees, Handshake AI 20+ on its license page and Scale AI prefers 40+, checked 2026-11-09.

How much do company data partnership programs pay?

Programs publish ranges as their own claims, and no company has publicly confirmed an amount received. Handshake AI shows two different maximums on one page, checked 2026-11-09.

Reading what company data is worth to AI companies means using dated sources, not headlines.

Is an email from a data partnership program legitimate?

Check that the sender's domain matches the program's own page and that the program publishes a legal entity. Alpheva AI's terms of service name no contracting entity, checked 2026-11-09.

The checks for whether a data licensing email is legitimate are set out separately.

Are these programs the same as AI training jobs?

No: company data programs license a business's records, while job listings recruit individual contributors.

Contributor work is compared under AI training jobs and platforms.