An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
- Yes. With a Global deployment, prompts and responses may be processed in any geography where the relevant model is deployed. That means processing can occur outside Japan.
- Not necessarily for stored data. For Global deployments, the documentation distinguishes between processing location and storage at rest:
- Processing/inferencing: may occur in any geography where the model is deployed.
- Data stored at rest: remains in the customer-designated geography.
- For storage behavior:
- For the core model inference path, the models are stateless: prompts and completions are not stored in the model.
- Prompts and completions are not used to train, retrain, or improve the base models.
- For Global deployments, if data is stored at rest, it remains in the customer-designated geography.
- Some Azure AI Foundry features are stateful and do store data in the service, such as the Files API, vector store, Responses API, Threads in the Assistants API, and Stored completions. For those features, stored data:
- is stored at rest in the Foundry resource in the customer’s Azure tenant, within the same geography as the resource,
- is encrypted at rest,
- can be deleted by the customer.
Applied to the questions:
- Can processing occur outside Japan? Yes, with Global deployment.
- Does that mean prompts/responses are stored outside Japan? No. The documentation says stored-at-rest data remains in the customer-designated geography.
- Is storage temporary or persistent?
- For normal stateless model inference, prompts and completions are not stored in the model.
- For specific stateful features, data is stored until deleted by the customer.
- Can stored data be outside Japan? For Global deployments, stored-at-rest data remains in the customer-designated geography. If the Foundry resource geography is not Japan, stored data follows that resource geography, while processing may still occur outside that geography.