Data protection query : Anthropic Claude via Azure AI Foundry (transfer chain & sub-processor status)

Question

We are completing a data protection transfer assessment for our use of Anthropic's Claude models accessed through the Azure AI Foundry model catalogue, and we need to confirm how personal data submitted to these models is protected under our agreement. Could you please confirm the following:

1. Which terms govern our use of Anthropic (Claude) models on Azure AI Foundry, the Microsoft Products and Services Data Protection Addendum (DPA), or separate "Non-Microsoft Product" / marketplace terms? If the latter, please point us to the applicable terms.
2. Is Anthropic engaged as a Microsoft sub-processor for model inference on the Foundry path (i.e. under Microsoft's DPA and Product Terms), or does Anthropic act under its own commercial terms and DPA for this processing? Is Anthropic listed on Microsoft's sub-processor list for the relevant Azure AI service?
3. Does the arrangement on the Foundry path mirror the sub-processor arrangement Microsoft applies to Anthropic in Microsoft 365 Copilot, or does it differ (for example, because Foundry catalogue models are treated as non-Microsoft products)?
4. Which SCC module applies to any transfer of personal data outside the EEA arising from this processing, and how is the flow-down of those obligations to Anthropic ensured?
5. For Anthropic (Claude) models on Foundry, where is model inference performed? Is there an EU data-residency / EU-region inference option for Anthropic models, and if not, are inference requests processed on Anthropic-operated infrastructure outside the EEA (including in the United States) regardless of the Azure region we configure?
6. What input/output retention applies to Anthropic models on this path (including any abuse-monitoring retention), and is a zero-data-retention or minimum-retention configuration available to us?

We would be grateful for a response, together with copies of (or links to) the applicable DPA and sub-processor documentation, so that we can retain them as evidence for our assessment.

Many thanks for your help.

Kind regards,

Angelika Churchman

Answer 1

Hello Angelika Churchman

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you are having issues with your Data protection query in Anthropic Claude via Azure AI Foundry (transfer chain & sub-processor status).

The issue is that Anthropic Claude in Microsoft Foundry must not be assessed as Azure OpenAI or as a Microsoft-owned model. Claude models in Microsoft Foundry are third-party Marketplace offerings from Anthropic, and Anthropic is the seller, operator, and independent data processor for prompts and outputs. - https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy, https://www.microsoft.com/licensing/terms/product/ForallOnlineServices/all

To resolve this, do the followings to:

• Classify Anthropic as an independent processor for Claude prompts and outputs in Microsoft Foundry, not as Microsoft’s Foundry inference subprocessor.
• Do not reuse the Microsoft 365 Copilot Anthropic subprocessor position for this Foundry scenario, because Microsoft 365 Copilot has a separate documented subprocessor arrangement. - https://learn.microsoft.com/en-us/microsoft-365/copilot/connect-to-ai-subprocessor, https://learn.microsoft.com/en-us/microsoft-365/copilot/copilot-anthropic-apps
• Confirm the selected Claude deployment type: Hosted on Azure or Hosted on Anthropic infrastructure.
• If Hosted on Azure, treat inference as processed on Azure infrastructure, with data at rest stored in the selected Azure geography and processing scoped to available Global/DataZone deployment options. - https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy
• If Hosted on Anthropic infrastructure, treat prompts and outputs as processed on Anthropic infrastructure, which may be outside Azure and outside the selected Azure region.
• Use Anthropic’s DPA/commercial terms accepted during deployment for Claude prompt/output processing and SCC validation.
• Do not assume Azure OpenAI retention or zero-data-retention controls apply to Claude. Retention and ZDR must be confirmed under Anthropic’s applicable arrangement for the exact model, feature, and organization. - https://platform.claude.com/docs/en/manage-claude/api-and-data-retention, https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
• If EU-only processing is mandatory, do not approve the deployment unless the exact Claude model/version and deployment option explicitly provide EU DataZone or EU-only processing commitments in official documentation and accepted terms.

After classifying Anthropic correctly, confirming the deployment hosting option, collecting the accepted Anthropic terms, validating the SCC route, and confirming retention/ZDR and regional commitments, the customer can complete the data-protection assessment accurately. If these deployment-specific details are not visible in the portal or public documentation, the next action should be to raise an Azure Support case via your Azure Portal or contact Priority Customer Support and request Product Group confirmation for the exact model, region, hosting option, retention, ZDR, and contractual position.

I hope this is helpful. Please! Do not hesitate to let me know if you have any other questions, steps or clarifications.

---

Please do not close the thread by upvoting and accepting the answer if any part of it is helpful.

Answer 2

Hi ,

Thanks for reaching out to Microsoft Q&A.

On Azure AI foundry, claude models are generally treated as non microsoft (3rd party) models offered via the model catalogue, so your usage is governed by Microsoft’s Product Terms + DPA for the Azure service layer, but model-specific terms (including data handling) may additionally fall under Anthropic’s own terms, which are surfaced through the marketplace/model catalogue listing; this means Anthropic is not typically positioned as a Microsoft subprocessor in the same way as core azure subprocessors, but rather as an independent data processor for inference when their model is invoked, unless explicitly stated otherwise in updated service documentation. This differs from M365 copilot scenarios, where microsoft tightly controls the processing boundary and may list providers differently under its sub-processor framework. For cross-border transfers, Microsoft’s DPA (including SCCs, typically Module 2/3 depending on controller/processor roles) applies to the azure layer, but flow-down to Anthropic depends on their own contractual commitments, so you need to review anthropic’s DPA to confirm alignment. Inference location and residency are not guaranteed to stay within the Azure region for 3rd party models; unless explicitly documented, requests may be processed on Anthropic-operated infrastructure (often US-based), meaning EU data residency may not be assured. Retention is also model-provider-specific: Microsoft enforces baseline protections (for example, no training on your data by default in enterprise contexts), but abuse monitoring and transient logging may still occur, and true zero-retention is not always available for third-party models unless explicitly stated. You should validate against the latest Azure AI Foundry model documentation, Microsoft DPA, and Anthropic’s trust/legal pages for audit evidence.

1. Do you require strict EU only processing with no cross-border fallback?
2. Are you treating microsoft as processor and Anthropic as subprocessor, or dual processors in your DPIA?

Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.