Microsoft Open AI Data Privacy Summary

Microsoft stores and processes data to provide generative AI services and to monitor for uses that violate their terms.

We are pleased to provide a summary of the data privacy details that users of Microsoft Azure need to understand in relation to Open AI services.

Hosting Locations

The Azure OpenAI service does not send your data to OpenAI. Microsoft hosts the OpenAI models within their infrastructure, and all customer data sent to Azure OpenAI stays inside the Azure cloud service. It is important to note that, currently, the Azure Open AI service can be hosted in Europe and the USA only. This is likely to change quickly and may already have by the time you read this post.

Data Retention

When you fine-tune an algorithm, you are sending training data to Microsoft. Training data is only used to fine-tune your model and is not used by Microsoft to train their own models.

When your users interact with a model (fine-tuned or not), they are submitting text prompts. Model inputs and outputs are processed through content filters as well as the specified OpenAI model. The Azure OpenAI Service stores both prompts and responses for up to thirty days for compliance and service monitoring purposes.

Data Access

Microsoft employees can see user interactions. If you have deployed your service within the EU, your data is only available to Microsoft employees located in the European Economic Area. If you would like to control who can view your information, you can request to modify content filtering and monitoring procedures in writing to Microsoft, which is subject to Limited Access eligibility criteria.

Occasionally, a Microsoft engineer may need access to your data to resolve service issues. In such cases, you can use the Azure Customer Lockbox, which enables you to approve or reject customer data access requests.

Feel free to contact us if you would like help to address the data privacy implications of Generative AI.