Trains on your data by default
= no (Favourable)are NOT used to train any generative AI foundation models without your permission or instruction.
service azure-openaivendor Microsoftcategory Inference hosts & routers
No (Favourable)
are NOT used to train any generative AI foundation models without your permission or instruction.
No prompts stored by default except for stateful features (stored in your resource's geography) and abuse-flagged samples kept for human review. Retention period for flagged samples is not stated on the page.
The models are stateless: no prompts or completions are stored in the model.
Against the 21 tools that state a period (grey dots):
Approved customers get no storage and no human review; automated review still runs.
If the customer has been approved for modified abuse monitoring
Sampled prompts are reviewed by authorized Microsoft employees. Data is stored in the resource's geography.
a sample of customer's prompts and completions may be selected for review.
'Global' deployments may process anywhere; 'DataZone' within the zone. APAC/Singapore availability not verified here.
prompts and responses may be processed in any geography where the relevant model sold by Azure is deployed
Microsoft lists in-scope services for ISO 27001 and SOC 2 in the 'Azure Compliance Offerings' PDF on the Service Trust Portal, which could not be downloaded without the portal. The public Learn pages do not name Azure OpenAI.
no official page stated this clearly
Microsoft offers a HIPAA BAA via its Product Terms for in-scope services. The in-scope list is in a Service Trust Portal PDF that could not be read, so Azure OpenAI's inclusion is not confirmed here.
no official page stated this clearly
Azure OpenAI models are hosted and operated by Azure. (Foundry Local, which runs other models on-device, is a separate product.)
Models sold by Azure are also hosted by Azure and operated by Azure as part of the Foundry Models service. They include all Azure OpenAI models
Spotted something wrong or out of date? Corrections are welcome; see how to send one. How statuses are assigned.