Vannus records that this vendor routes across several model providers — not captured by one. Every finding below is quoted to the vendor’s own document, or marked not disclosed where the vendor publishes nothing.
Microsoft's enterprise AI assistant embedded across Microsoft 365 apps, grounding generative responses in the customer's own Microsoft Graph data — email, documents, chat, calendar and meetings.
Microsoft offers third-party AI models in Microsoft Copilot, such as Anthropic and OpenAI models.
This page gives the finding and links the document it comes from. The AI Subprocessor Jurisdiction Report ($299) quotes the clause this finding rests on, and the clause for each other vendor in your stack where the report can quote one, in one dated document you can hand to an auditor.
Applies to: Microsoft 365 Copilot (including Microsoft 365 Copilot Search) for Microsoft 365 commercial customers
From the product security/privacy documentation (Microsoft Learn): “Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs, including those used by Microsoft 365 Copilot.” learn.microsoft.com ↗
Training and retention posture varies by plan. The default-plan row above describes the vendor’s free or standard tier; the paid-plan row is quoted from the document linked beside it. A negotiated contract can override either. Check your own agreement before relying on this.
On U.S. CLOUD Act reach specifically: the statute reaches a provider subject to U.S. jurisdiction over data in its possession, custody or control. Corporate control is a strong indicator of that and it is what we can evidence from published documents — but it is not the whole test. A company founded outside the U.S. can still contract through a U.S. entity or run substantial U.S. operations. Treat this as a starting point for your own review, not a legal determination, and take advice on anything that matters.
This is the history of our record of this vendor: when we first recorded it; each time one of these changed: the determination sentence, the contracting entity, its country, the governing law, the US-control finding, what the answer depends on, the note beside the finding, or the cited document; and each time the vendor's document was read again and none of them changed. A change only in capitals, punctuation or spacing is not a line, and neither is a first document title. A line names what changed and does not repeat what those fields said before; the determination above says what they say now. Where the determination sentence itself changed, the line shows the earlier sentence too, marked as before, with why it changed. The automated re-check's latest result is not a line here; it is printed beside the clause above. A change here may be the vendor changing its document or us correcting our reading of it, and a line says which only where the record shows it; otherwise it says the cause is not established. Each line is today's reading of the record as it stood on that date. A reading we later found its citation does not carry was withdrawn, and is not repeated here. This history as data.
The vendor's published or catalog-recorded posture — the concrete facts this entry is built from. A full audit verifies each against the vendor's current documentation.
Microsoft 365 Copilot: “Contracting entity depends on your region or order form — read the clause.” Vannus, https://vannus.co/tool/microsoft-365-copilot microsoft.com · read 2026-09-21
Vannus records, from the vendor's own published documents, the legal entity a customer contracts with, the country that entity sits in, and the governing law of its terms — and whether the vendor runs its own model or resells someone else's. Each finding is quoted to its source and dated, or marked not disclosed where the vendor publishes nothing. No paid placement on a record, in the catalog or in search — affiliate status is walled off from the record, enforced by a test in the build. See the methodology →