Can you license company emails to AI?
The short answer
Yes, if counsel confirms you hold the rights. Troveo lists whole email threads. Google bid $10 million for Spirit's 100 million emails and more. No per-email price is public. Law firm Frankfurt Kurnit advises excluding privileged and HR mail.
Email shows who asked for what, who approved it and how it ended. micro1's CFO says labs want that chain.
It is also full of other people's words. Employees, customers and partners all write into the same archive, which is why the rights work comes before any price talk.
Which companies take company emails?
The companies list email among the systems they take (from their pages, plus one talk):
- Troveo lists "Email & calendar" as a category: threads with senders, recipients and linked responses.
- Mercor lists Gmail ("email messages and metadata") and Outlook among 100+ connectors.
- micro1's CFO, in an undated recorded forum talk on its company data partnerships, described the chain from a request to a deliverable across tools. Email is one link next to chat and documents.
I have no public link to that recording, so read the CFO's remarks on this page as micro1's own account.
None of the three names an end buyer for email. My page on who bid on Spirit Airlines' data lists the named bidders.
What do Troveo and Mercor list for an email archive?
Troveo lists whole email threads with senders, recipients and linked responses, and Mercor lists Gmail messages with their metadata. One message on its own shows little. The reply chain, who was copied and the order of the answers show how the work moved.
So plan any export to keep headers and thread links.
What does a lab do with company email?
Time reported on 25 Aug 2026 that AI companies build reinforcement learning environments. These are simulations of commonly used software where agents practice tasks.
Nick Heiner of Surge AI told Time that RL environments are only as good as the data that populates them. The article adds that training agents for other office work may depend on private material held in company email and chat.
Epoch AI's review of RL environments (12 Jan 2026) describes app replicas and lab partnerships with product companies, not real company archives. Demand for archives is a claim from buyers and press, so treat it as one.
What is an email archive worth?
No public price per email exists, and none of the figures here estimates what your archive would fetch. The public bids are for one whole archive:
- Google won Spirit Airlines' bankruptcy auction with a $10 million bid for about 100 million emails, 500 million Teams messages and about 30 million lines of code (Time).
- Mercor bid $7.5 million (Time) and is the alternate bidder (ombudsman's report).
- micro1 later proposed $12.5 million for the same records, Fortune reported on 14 Sep 2026.
micro1 also states its own range for company data. Its Enterprise Data Partnership listing says $100K-$2M+ for approved data packages (self-reported).
Its CFO gave $100,000 to $2 million per company in a separate, undated recorded talk. Neither is a price for email alone.
Spirit was one distressed airline's whole archive, and the sale is still waiting on the court (docket):
Google's $10M winning bid for Spirit Airlines' de-identified business data still needs bankruptcy court approval. A court notice reported on 9 Oct 2026 moved the hearing from 14 Oct to 6 Nov 2026. No sale order as of 10 Oct 2026.
Read the bid as one data point. My Spirit page has the timeline.
Do rights or price come first?
Rights. An archive holds many kinds of messages:
- Employees' messages, which raise privacy and labor questions.
- Customers' messages.
- Partners' messages under NDAs.
- Sometimes legal advice covered by attorney-client privilege.
Owning the mailbox and having the right to license what is inside it are separate questions. Who owns company emails pulls them apart.
What has to come out first?
Daniel Goldberg, a partner at Frankfurt Kurnit, writing in October 2026, advises owners to exclude high-risk data entirely instead of relying on de-identification. His list:
- Privileged material.
- HR and personnel files.
- Payroll and tax records.
- Health and benefits information.
- Union-related communications.
- Customer and partner confidential information.
- Licensed data and data from regulated systems.
He warns that licensing may waive privilege over legal communications. Employee messages about pay, scheduling and working conditions may be protected concerted activity under Section 7 of the National Labor Relations Act, with or without a union.
He also notes that removing identifying details from millions of emails is harder than scrubbing names from structured fields.
He adds that de-identification deals with who can be linked and leaves the content as confidential as it was. A disciplinary thread stays sensitive without a name on it.
Who removes the names?
It varies by deal. micro1 staff said in the undated forum talk that micro1 transforms the data itself. The same talk said companies cannot scrub first, because redaction would break the cross-tool trail.
Troveo's page says names, contact details, credentials and client identities are scrubbed, and its example swaps them for labels such as PERSON-014 and EMAIL-019.
In the Spirit sale the debtors hired an independent agent, Tonic.ai, to de-identify unstructured data such as corporate emails, the court's consumer privacy ombudsman reported in an October 2026 supplemental filing.
Before each transfer, Tonic.ai is to deliver a certification, filed with the court, that the approved procedures ran.
That certified de-identification is a closing condition of the sale. The ombudsman wrote that she did not investigate whether the sale presents risks to affected employees or third parties.
She also noted that micro1 proposed to do the de-identification itself, without an independent party, and asked for time to look into that if the court considers micro1's bid.
Stand-ins that stay the same across messages keep the links between them, and those links are also the re-identification risk. My re-identification page and the de-identification explainer cover it.
What does an email mailbox inventory look like?
The first job is a list of what you have, in descriptions and counts only. Daniel Goldberg, a partner at Frankfurt Kurnit, suggests listing each data set by system, record count and start date. Retention rules and who runs each system belong on the same sheet.
The example on this page is a made-up sheet for Larkspur Freight, a 60-person freight company.
Fictional mailbox inventory for Larkspur Freight, a made-up 60-person freight company. Counts and descriptions only, no messages.
| Mailbox group | Note for counsel | Messages, rough | Retention or hold | Who runs it |
|---|---|---|---|---|
| Sales, 14 accounts, 2016-2025 | Customer pricing under NDAs, check NDAs | 1.9 million | Kept 10 years by policy | Outside IT firm |
| Dispatch, 22 accounts, 2018-2025 | Drivers' personal details | 2.6 million | No policy, nothing deleted | Office manager |
| Shared inboxes (info@, support@), 5 accounts, 2022-2025 | Customers' own messages | 1.2 million | 3-year auto-delete since 2023 | Support lead |
| Former staff, 31 accounts, 2016-2024 | Departed staff's personal mail | 800,000 | Archived as PST files | Outside IT firm |
| Finance and HR, 4 accounts, 2016-2025 | Payroll and health, exclude | 310,000 | Litigation hold since 2021 | CFO |
| Executive and legal, 3 accounts, 2016-2025 | Privileged advice, exclude | 240,000 | Litigation hold since 2021 | CEO |
A sheet like this lets you talk scope with a buyer before one message leaves the building.
In this example:
- Finance, HR and the executive mailboxes hold categories from Goldberg's list, so on his advice they come out of scope before any buyer sees the sheet.
- Sales holds customer pricing under NDAs, also on his list. Ask counsel whether those NDAs allow any use at all.
- Two groups sit under a litigation hold. Ask counsel whether held mail can move at all.
- The shared inboxes have deleted anything older than three years since 2023, so older support threads are gone. A buyer that wants whole threads will ask about that gap.
Larkspur, being imaginary, has a lot less paperwork than you.
How does an email deal run?
Start with the rights check, which walks through seven rights questions, including staff messages, customer contracts, privilege, NDAs and who can sign. Then read the risk map.
Optional, paid managed readiness builds the inventory with you, quoted in writing first. We work from inventories, counts and file lists, and never accept raw records, credentials or data-room access.
micro1 may pay us a referral fee if a company we introduce goes ahead with it. How we are paid
Apply to micro1's data partnership (paid link)micro1's own data-partnerships pagePaid link. How we are paid
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. Ask for a referral
