Does Azure OpenAI / Microsoft Foundry (Azure AI Foundry) sign a BAA?
Not confirmed
Microsoft's HIPAA BAA is included automatically, through the Product Terms and DPA, for in-scope Azure services. However, no Microsoft page we could reach names Azure OpenAI or Foundry models as in scope; the official list sits in a Service Trust Portal PDF we could not open. Treat coverage as unconfirmed until verified there or with Microsoft.
- Business associate
- Microsoft
- Tool
- Azure OpenAI / Microsoft Foundry (Azure AI Foundry)
- Plans with a BAA
- Not stated
- Checked
By plan
| Plan | BAA | How to get it | Conditions |
|---|---|---|---|
| Azure subscription (EA / MCA / CSP), Models sold by Azure incl. Azure OpenAI | Not confirmed | Included in the terms | The Microsoft BAA comes by default through the Product Terms and DPA, with no separate signature, but only for in-scope Azure services. Whether Azure OpenAI / Foundry Models sold by Azure are in scope could not be confirmed from an official page. Microsoft Q&A moderators (not official documentation) give conflicting answers.[2],[3],[4] |
| Foundry Models from partners and community (serverless / Marketplace) | Not confirmed | Not stated | Microsoft treats these as Non-Microsoft Products governed by each provider's terms and billed through Azure Marketplace. No official HIPAA statement for them was found.[5] |
Are the AI features covered?
Vendors often sign a BAA for the core product but leave some AI features out. Here is what Microsoft says about each one.
| AI feature | Under the BAA | Details |
|---|---|---|
| Azure OpenAI models (Models sold by Azure)Azure subscription (EA / MCA / CSP), Models sold by Azure incl. Azure OpenAI | Not stated | No reachable official Microsoft page lists Azure OpenAI in HIPAA BAA scope. The data-privacy doc is silent on HIPAA.[1],[4] |
| Abuse monitoring / modified abuse monitoringAzure subscription (EA / MCA / CSP), Models sold by Azure incl. Azure OpenAI | Not stated | By default, flagged prompts and completions may be stored for human review. Turning that off (modified abuse monitoring) is limited to managed customers who meet extra criteria. Microsoft docs don't link this to HIPAA.[1],[6] |
| Claude models in Microsoft Foundry (under Anthropic's BAA)Foundry Models from partners and community (serverless / Marketplace) | Not stated | Anthropic's pages conflict. Its BAA article lists Claude via Foundry as eligible under its BAA, while its API docs say HIPAA readiness isn't available on Microsoft Foundry.[7],[8] |
| Other partner/community models (e.g., Mistral, DeepSeek, Qwen)Foundry Models from partners and community (serverless / Marketplace) | Not stated | No official Microsoft HIPAA statement found for individual partner models.[5] |
Cloud AI platforms that offer a BAA
- SupabaseBAA on: Team / Enterprise with HIPAA add-onSome plans only
- Amazon BedrockBAA on: AWS accounts (standard AWS BAA)Signs a BAA
- AWS HealthScribeBAA on: AWS accounts with the AWS BAA accepted (US East, N. Virginia)Signs a BAA
- Google Cloud Vertex AI (Gemini Enterprise Agent Platform)BAA on: Google Cloud customers (Agent Platform / Vertex AI, GA services)Signs a BAA
All BAA-ready alternatives to Azure OpenAI / Microsoft Foundry (Azure AI Foundry)
Compare Azure OpenAI / Microsoft Foundry (Azure AI Foundry)
- Azure OpenAI vs Amazon BedrockAmazon Bedrock: signs a BAA
- Azure OpenAI vs Google Cloud Vertex AIGoogle Cloud Vertex AI: signs a BAA
Sources
- Data, privacy, and security for Models sold by Azure
- HIPAA (US) Health Insurance Portability and Accountability Act - Azure compliance
- HIPAA/HITECH - Microsoft compliance
- Cloud services in audit scope
- Foundry Models overview
- Limited access to Models sold by Azure
- Business Associate Agreements (BAA) for commercial customers
- API and data retention
A signed BAA is necessary but not sufficient for HIPAA compliance, which also depends on configuration, access controls, and your own policies. Verdicts summarize the vendor's own published terms as of the date checked; confirm with the vendor and your counsel before sending protected health information. Not legal advice.