Azure OpenAI / Foundry Models sold by Azure
From AI API data terms: training, retention and zero retention: What does each AI provider say happens to data sent to its commercial API: is it used for training, how long is it kept, and can zero retention be had?
Provider: Microsoft Azure
The company offering the API, as named on its own page.
Foundry is an Azure service; Microsoft hosts the Models sold by Azure in Microsoft's Azure environment
Source: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy. Read 2026-11-01. Checked by a second reader.
Offering: Foundry Models sold by Azure
The API, tier or platform the row covers, in the provider's words.
Foundry Models sold by Azure detect and mitigate instances of recurring content and/or behaviors
Source: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/abuse-monitoring. Read 2026-11-01. Checked by a second reader.
Trains on API data by default: No
True or false only where the text plainly says what happens to this offering's API data with no opt-in or opt-out by the customer. An opt-out with no stated default is left blank. Where pages differ: legal or privacy page first, then help or trust pages, then developer docs, with the other in conflicts.
are NOT used to train any generative AI foundation models without your permission or instruction.
Source: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy. Read 2026-11-01. Checked by a second reader.
Default retention: Not found
How long prompts and outputs are kept by default, in the provider's words. Where pages differ: legal or privacy page first, then help or trust pages, then developer docs, with the other in conflicts. A page that's only less specific isn't a conflict: the more specific statement is the value.
No source we could use stated this. A blank never means no.
Exceptions: Azure preview features, including models sold by Azure in preview, may employ different privacy practices, including with respect to abuse monitoring
The main case with different retention, in the provider's words: first a model carved out of the general promise, then a feature, then a case such as flagged content. Pages rank as for Default retention; between pages of equal rank, the one naming the models. One passage only, so others may exist.
Azure Preview features, including Models sold by Azure in preview, may employ different privacy practices, including with respect to abuse monitoring.
Source: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy. Read 2026-11-01. Read once.
Zero retention offered: Not found
True only where the provider plainly offers zero data retention by that name for this offering; false only where it plainly excludes it. A similar control under another name is left blank here and described in How to get it.
No source we could use stated this. A blank never means no.
How to get it: Customers who meet additional limited access eligibility criteria apply to modify abuse monitoring by completing this form; some advanced models may have more stringent criteria
What the provider says is needed for zero retention. Where it has no offer by that name, its nearest named control or pointer, in its own words. Where pages differ: legal or privacy page first, then help or trust pages, then developer docs, with the other in conflicts.
Microsoft allows customers who meet additional Limited Access eligibility criteria to apply to modify abuse monitoring by completing this form. Some advanced Models sold by Azure may have more stringent criteria for turning off abuse monitoring.
Source: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/abuse-monitoring. Read 2026-11-01. Checked by a second reader.
BAA available: Not found
True where the provider says it signs or offers a HIPAA Business Associate Agreement, or sets terms for use under one, for this offering. A general cloud BAA that the page doesn't tie to this offering, or an invitation to ask, is left blank.
No source we could use stated this. A blank never means no.
Data location: Processed within the customer-specified geography (unless you are using a global or datazone deployment type)
Where the provider states data is processed or stored, in its words. Where pages differ: legal or privacy page first, then help or trust pages, then developer docs, with the other in conflicts.
Prompts and responses are processed within the customer-specified geography (unless you are using a Global or DataZone deployment type), but may be processed between regions within the geography for operational purposes (including performance and capacity management).
Source: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy. Read 2026-11-01. Checked by a second reader.