| Quick facts | |
|---|---|
| API terms on AI training | No express AI clause found[1] |
| Does Microsoft 365 & Teams train on your data? | Vendor states it does not train on content[2] |
| Recommended export path | Microsoft Purview eDiscovery searches and exports mailbox, Teams and SharePoint content for a case.[3] |
| Connector risk | Lower on the API terms we read — but mailbox content belongs partly to every correspondent, which is the larger issue. |
Section 1
What Microsoft 365 & Teams holds that AI buyers value
- Exchange mailboxes
- Teams chats and channel messages
- SharePoint and OneDrive documents
Where the value is: Email and document archives show how work moves between people and files over years.
Section 2
How to export your Microsoft 365 & Teams data
Microsoft Purview eDiscovery searches and exports mailbox, Teams and SharePoint content for a case.[3]
- Create an eDiscovery case scoped to the custodians and date range in question.
- Search, review counts, and export.
- Keep the search query and export report with the data.
Keep a note of who ran each export, when, and with what filters. It is the first page of the documentation file a buyer — and later an acquirer — will ask for.
Section 3
What Microsoft 365 & Teams’s API terms say about AI
The Microsoft APIs Terms of Use (October 2025) contain no express AI-training restriction that we found.[1]
Absence of an express clause is not permission. Your subscription agreement, acceptable-use policy and order form still govern what you may do with records held in the platform, and other vendors have added AI clauses during 2025–2026. Use an administrator export, and have counsel read the current agreement.
Section 4
Does Microsoft 365 & Teams train AI on your data?
Microsoft has denied using customer data from Microsoft 365 applications to train AI models.[2]
A vendor’s own use of your data does not transfer ownership, but it belongs in your disclosure to a buyer: it bears on uniqueness and on whether the same records have already fed a competing product.
Section 5
Fields to scrub before licensing
- Every email address and signature
- Third-party correspondence
- Privileged communications with counsel
- HR files
Removing names is not de-identification; combinations of quasi-identifiers re-identify people.[4] California’s statute sets the bar for a “deidentified” consumer record.[5] Test a sample with the browser-local PII scanner before anyone else sees it.
Section 6
Check the rest of your stack
Most licensable files span several systems. The software AI data rights checker lists the API clause, vendor-training position and export path for every system we have reviewed, and the SaaS AI data ownership index publishes the same findings as a downloadable dataset.
Frequently asked questions
Can I sell my Microsoft 365 & Teams data to AI companies?
How do I export all my data from Microsoft 365 & Teams?
Does Microsoft 365 & Teams use my data to train AI?
References
Sourced from primary industry reports, government data, SEC filings, and peer-reviewed valuation research. Retrieved at time of publication.
- 1.Microsoft Corporation — Microsoft APIs Terms of Use — Last updated October 2025; we found no express AI-training restriction in this document (2025)
- 2.Reuters — Microsoft denies training AI models on user data — Microsoft denied using customer data from Microsoft 365 applications to train AI models; 27 November 2024 (2024)
- 3.Microsoft Corporation — Export search results in eDiscovery — Microsoft Purview eDiscovery export of mailbox, Teams and SharePoint content
- 4.U.S. National Institute of Standards and Technology — NISTIR 8053 — De-Identification of Personal Information — Reference treatment of de-identification techniques and re-identification risk (2015)
- 5.State of California — California Consumer Privacy Act, Civil Code § 1798.140 — definitions of "deidentified" and "sale" — Statutory standard a de-identified operational record must meet, and the breadth of "sale"