Antrophic models sold in Foundry Marketplace

Hello @Remus Tanasa

Thank you for reaching out Microsoft Q&A,

Data usage and retention (Claude via Foundry)

When using Anthropic Claude models through Azure AI Foundry Prompts and outputs are not used for model training by Anthropic by default.

Anthropic does not use customer prompts or completions from commercial/API usage (including Foundry) to train their foundation models

Training would only occur if there is an explicit opt-in scenario

Data processing and retention

It’s important to distinguish training from processing Prompts and outputs are processed and may be retained by Anthropic, as they are the model provider.

Key points:

Data processor

Anthropic acts as the data processor for Claude models

Model inference is performed on Anthropic-managed infrastructure

Retention

Data may be retained for:

• Abuse detection
• Safety monitoring
• Service reliability

Retention is typically short-term (e.g., ~7–30 days)

Content flagged for policy or abuse review may be retained longer

Scope

Azure OpenAI “zero data retention” guarantees do not apply to Anthropic models

Microsoft vs Anthropic responsibilities

Microsoft

• Provides the Foundry platform, deployment, and billing layer
• Does not use your data for training

Anthropic

• Hosts and runs the Claude models
• Governs data processing and retention policies

Anthropic models in Foundry are offered as third-party SaaS models, so:

• Azure documentation may not explicitly detail all data handling behavior
• Final authority on processing/retention lies with Anthropic’s policies and Data Processing Addendum

Please refer this

Foundry Models from partners and community – Anthropic https://learn.microsoft.com/azure/foundry/foundry-models/concepts/models-from-partners?wt.mc_id=azure-cxp-community-insider#anthropic

Data, privacy, and security for use of models through the Model Catalog https://learn.microsoft.com/azure/machine-learning/concept-data-privacy?view=azureml-api-2

 I Hope this helps. Do let me know if you have any further queries.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thank you!

Yep - prompts and outputs submitted to Claude models via Microsoft Foundry are not used for model training by Anthropic. Anthropic does not use data from their commercial products (including the API deployed in Foundry) to improve their foundation models unless you explicitly opt in.

While data is not used for training, it is retained and processed by Anthropic on their infrastructure for operational and safety purposes. For Claude in Foundry, Anthropic acts as the data processor, meaning your prompts and completions are sent to Anthropic-managed servers rather than remaining entirely within Microsoft’s local environment. Standard API retention typically lasts for a short period (such as 7 to 30 days) to support service functionality and abuse monitoring, though sessions flagged for policy violations may be held for up to two years.

In this deployment model, Microsoft manages the billing and API endpoint infrastructure, but the actual "thinking" (inference) is performed by Anthropic-hosted services. Because Anthropic governs the processing of these prompts and responses, their specific commercial terms and Data Processing Addendum apply to how that information is handled.

For detailed confirmation, you can refer to Microsoft documentation on data privacy for Claude in Foundry.

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