AI Assistant Privacy FAQ¶
Amperity’s AI Assistant is powered by GPT-4, which runs on Azure’s OpenAI Service.
Azure OpenAI Service has the following data sharing policies:
- No data is available to or shared with other customers
- No data is available directly to OpenAI
- No data is used to improve OpenAI models
- No data is used to improve any Microsoft or 3rd party products or services
What data does the Azure OpenAI Service store?¶
Azure OpenAI Service does not store any data after the API response is returned. The GPT-4 model is stateless and all context must be sent with each API call.
What data does Amperity store?¶
Amperity stores conversation history in the same secure location as existing tenant data. This data is isolated per tenant and is never aggregated or intermixed between or across tenants.
What data does the Azure OpenAI Service process?¶
There are two types of data Amperity sends to the Azure OpenAI Service:
- Data to improve model performance:
- Database schema information including tables names, fields and description
- The current query’s SQL
- Sample of results from the query (if available)
- Any errors encountered
- AI Assistant message history: Amperity sends the full message history for a given conversation to Azure OpenAI Service to maintain the context of the conversation.
What data is not sent to Azure OpenAI Service?¶
Amperity sends only the data needed to ensure effective operation of the Azure OpenAI Service and no data is stored at rest in the Azure OpenAI service. In addition, Amperity restricts the use of the service to limit access to data:
- Extensive customer PII: Amperity will send a sample of query results to the Azure OpenAI Service which may contain PII. The sample query results are capped at ten rows of data.
- Credentials: Amperity does not send any stored upstream or downstream credentials (usernames and passwords) to Azure OpenAI Service.