An Azure offering that provides a suite of purpose-built technologies for protected health information in the cloud.
These are the steps you could follow to answer your questions in details:
- Contract/BAA proof Your BAA for Azure OpenAI comes via Microsoft’s DPA, not a per‑service contract; download and file it with your EA/MCA/CSP. Docs: DPA (Microsoft BAA terms) here: https://www.microsoft.com/licensing/docs/view/Microsoft-Products-and-Services-Data-Protection-Addendum-DPA; HIPAA offering/roles: https://learn.microsoft.com/en-us/azure/compliance/offerings/offering-hipaa-us. Confirm with your account team that Azure OpenAI is consumed under that agreement.
- Scope discipline (text only for PHI) Until Microsoft explicitly lists modality coverage, treat PHI as text‑only in Azure OpenAI; don’t send PHI via images or realtime audio. Image scope caution (community moderator) → https://learn.microsoft.com/en-us/answers/questions/2106637/azure-openai-hipaa-compliance-status. Realtime audio uncertainty (eligibility not confirmed) → https://learn.microsoft.com/en-us/answers/questions/5616040/clarification-request-hipaa-eligibility-of-azure-o.
- Private network boundary Place Azure OpenAI behind a VNET + Private Endpoints (no public ingress), and front with your private API gateway if needed. Azure OpenAI FAQ confirms VNET/Private Endpoints support here: https://learn.microsoft.com/en-us/azure/ai-foundry/openai/faq?view=foundry-classic. Keep architecture diagrams and endpoint screenshots as audit evidence.
- Identity, access, and encryption Enforce MFA/Conditional Access and least‑privilege RBAC via Microsoft Entra ID; require TLS 1.2+ and encrypt any stored data. This maps to HIPAA’s administrative/technical safeguards under shared responsibility: https://learn.microsoft.com/en-us/azure/compliance/offerings/offering-hipaa-us. Document policies, control owners, and monitoring in your evidence pack.
- Data‑handling posture (no model training on your data) Azure OpenAI (Azure Direct Models) does not train foundation models on your prompts/outputs without your instruction/permission. Microsoft’s data/privacy statement for Azure Direct Models: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy?view=foundry-classic. Keep this page in your auditor binder and reference it in your DPIA/Risk Analysis.
- Abuse‑monitoring retention decides and verify Decide whether to allow Microsoft’s abuse‑monitoring storage; if you require stricter control, verify it’s off and retain proof. How to verify storage for abuse monitoring is off: https://learn.microsoft.com/en-us/azure/ai-foundry/responsible-ai/openai/data-privacy?view=foundry-classic#how-can-a-customer-verify-if-data-storage-for-abuse-monitoring-is-off. Store screenshots/change tickets alongside your logging/redaction policy.
- If using OpenAI (direct) instead of Azure For OpenAI API (non‑Azure), obtain a separate BAA from OpenAI; ChatGPT consumer tiers are not HIPAA‑eligible. OpenAI BAA request instructions: https://help.openai.com/en/articles/8660679-how-can-i-get-a-business-associate-agreement-baa-with-openai-for-the-api-services. Regardless of path, maintain an evidence bundle: signed master agreement, current DPA, configs (VNET/PE), and data‑handling records.
If the above response helps answer your question, remember to "Accept Answer" so that others in the community facing similar issues can easily find the solution. Your contribution is highly appreciated.
hth
Marcin