Short answerMicrosoft states that prompts, completions, embeddings and training data in its Azure-sold models are not available to other customers or to OpenAI and are not used to improve models. It also lists caveats such as abuse monitoring and some cross-region processing.
- Microsoft says your prompts, completions, embeddings and training data are not available to other customers or to OpenAI, and are not used to train foundation models without your permission.
- It says the models are hosted in Microsoft's Azure environment and do not interact with services operated by the model providers, such as ChatGPT or the OpenAI API.
Caveats Microsoft lists
- Abuse monitoring may store and sample prompts and completions for review; a modified monitoring option is available by application.
- Some deployment types may process data in other geographies.
- Features such as the Responses API persist data in your tenant, encrypted, and deletable.
It is a custom build, not a one-click migration. Use the privacy checklist and see self-hosted vs cloud.
Not sure it is simple enough to do yourself?
Tell us what your GPT does, whether it uses Actions and how many files it has. We will say what will break and what a rebuild involves. The audit is free.
Request a free auditRelated guides
Sources
Pages describing other companies' products reflect their public documentation on the date shown and can go out of date.