Does Microsoft 365 Copilot train on your data?
Microsoft · Workplace AI suite · official site ↗
Doesn’t train on your dataZDR / DPA on enterpriseNo training to opt out of
Does Microsoft 365 Copilot train its models on your data?
No — Microsoft 365 Copilot does not use your content to train its AI models by default.
Microsoft does not use Microsoft 365 Copilot prompts, responses or Microsoft Graph data to train the underlying foundation models; usage is covered by your existing commercial terms and Enterprise Data Protection.
Can you opt out?
There is no training opt-out to set: prompts and responses are not used to train foundation LLMs at all. Note this covers commercial Microsoft 365 Copilot — the free consumer Microsoft Copilot app runs under different consumer terms.
Zero retention / DPA
Prompts/responses are protected by the same DPA and contractual commitments as Exchange and SharePoint, and Microsoft 365 Copilot is an EU Data Boundary service. Caveat: Anthropic models offered in Copilot are out of scope for the EU Data Boundary, and flex routing can process inference outside the EU under peak load. DPA ↗
What the listicles get wrong
Strong default, but the EU Data Boundary has documented carve-outs: Anthropic models are out of scope and flex routing can move inference outside the EU. Scope here is commercial M365 Copilot, not the consumer Copilot app.
Verdict by plan tier
Microsoft 365 Copilot (licensed/commercial)No trainingPrompts, responses and Graph data are not used to train foundation models; covered by Enterprise Data Protection and the commercial DPA.
Microsoft 365 Copilot Chat (Entra-authenticated)No trainingSame enterprise terms and protections; not used for foundation-model training.
Get notified when this changes
We track Microsoft 365 Copilot's data-training and retention policy. Leave your email and we'll send one note if it changes.
Frequently asked questions
Does Microsoft 365 Copilot train its AI models on my data?
No — Microsoft 365 Copilot does not use your content to train its AI models by default. Microsoft does not use Microsoft 365 Copilot prompts, responses or Microsoft Graph data to train the underlying foundation models; usage is covered by your existing commercial terms and Enterprise Data Protection.
Can I opt out of Microsoft 365 Copilot training on my data?
There is no training opt-out to set for Microsoft 365 Copilot: There is no training opt-out to set: prompts and responses are not used to train foundation LLMs at all. Note this covers commercial Microsoft 365 Copilot — the free consumer Microsoft Copilot app runs under different consumer terms.
Does Microsoft 365 Copilot offer zero data retention (ZDR) or a DPA?
Prompts/responses are protected by the same DPA and contractual commitments as Exchange and SharePoint, and Microsoft 365 Copilot is an EU Data Boundary service. Caveat: Anthropic models offered in Copilot are out of scope for the EU Data Boundary, and flex routing can process inference outside the EU under peak load.
Is Microsoft 365 Copilot safe to use with confidential or proprietary data?
It depends on your plan tier. Microsoft 365 Copilot (licensed/commercial): Prompts, responses and Graph data are not used to train foundation models; covered by Enterprise Data Protection and the commercial DPA. Microsoft 365 Copilot Chat (Entra-authenticated): Same enterprise terms and protections; not used for foundation-model training. Always confirm current terms with Microsoft before sending confidential data — this is cited public information, not legal advice.
Sources
https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy
https://learn.microsoft.com/en-us/microsoft-365/copilot/enterprise-data-protection
https://learn.microsoft.com/en-us/privacy/eudb/eu-data-boundary-excluded-services
This page is cited public information, not legal or compliance advice. Whether Microsoft 365 Copilot trains on your data, and any zero-retention or DPA option, can depend on your specific plan, region and contract. Always confirm current terms with Microsoft before sending confidential or proprietary data.