AI Data Privacy and Residency
Outermind runs its AI agents on models delivered through Microsoft Azure AI Foundry. A question we hear often is: when Outermind uses a model like OpenAI's GPT or Anthropic's Claude, does my data (my prompts, my documents, the model's outputs) get sent to those AI companies?
This page answers that question using Microsoft's and Anthropic's own published documentation. Every factual claim below links to its source. Because these platforms change frequently, the linked Microsoft tables are always the authoritative, live source; the snapshots on this page are dated so you know how current they are.
Two terms to know first
Before the details, two terms that get confused:
- Data at rest is your data as it is stored, for example when it sits in a database or a storage account. Data residency for data at rest is about which country or region the data is kept in.
- Inference processing is the short-lived work of running your prompt through the model to produce an output. During inference your prompt and the response pass through compute (including GPUs), but they are being processed, not permanently stored. Where inference happens can differ from where data at rest is kept.
Keeping these separate matters: a deployment can store your data at rest in one region while still, depending on its configuration, processing inference in another.
OpenAI models on Azure ("Models sold by Azure")
OpenAI models offered through Azure AI Foundry are in the category Microsoft calls "Models sold by Azure." These are hosted by Microsoft in Azure. According to Microsoft's data, privacy, and security documentation for these models:
- Your prompts, completions, and uploaded data are not available to the model provider (for example, OpenAI).
- They are not used by providers to improve their models.
- They are not used to train foundation models without your permission.
Microsoft is explicit that these models "do NOT interact with any services operated by providers of Models sold by Azure, for example, OpenAI (e.g. ChatGPT, or the OpenAI API)." (See the same Microsoft data-privacy page.) In other words, using an OpenAI model through Azure does not route your data to OpenAI's own consumer or API services.
Anthropic Claude models: two hosting options
Claude models on Azure AI Foundry can be deployed under one of two hosting options, and the difference matters for where your data is processed. Microsoft documents this in its Claude models data-privacy page and its Claude hosting comparison.
- Hosted on Azure is the default. Per the Microsoft guide for using Claude models, deployments created with Default settings are "automatically set to version 2: Hosted on Azure." With this option, prompts and outputs are "processed on Azure infrastructure, including request ingress, API services, and GPU inference," and data at rest stays in the selected Azure geography.
- Hosted on Anthropic infrastructure processes prompts and outputs on Anthropic-hosted infrastructure. Under this option, Microsoft states that "data might be processed outside of Azure including outside of your selected Azure region."
So the hosting option chosen for a Claude deployment determines whether inference stays on Azure infrastructure or runs on Anthropic-operated infrastructure.
Anthropic's role (applies to both hosting options)
Regardless of hosting option, Anthropic is the seller, operator, and contractual data processor for Claude prompts and outputs. This relationship is governed by Anthropic's Data Processing Addendum and Commercial Terms of Service.
- Under the Commercial Terms, Anthropic does not train its models on customer content.
- There is a narrow safety exception: automatic safeguards can flag content, and Anthropic personnel review flagged content on an exceptions-only basis for potential safety violations.
For more detail on Anthropic's controls and certifications, see the Anthropic Trust Center.
Data residency: where processing and storage happen
Data residency depends on both the model category and the specific deployment type. Keep the data-at-rest vs inference-processing distinction above in mind: the guarantees differ between the two.
Claude models
For Claude, the only geo-fenced inference option is Data Zone Standard (US). In Microsoft's region availability tables, the Europe and Asia Pacific Data Zone tables for Claude show "Not available." All other Claude deployments are Global Standard, which Microsoft defines as processing "in any Azure region where you deploy the model." Data at rest for Claude remains in the selected geography.
Models sold by Azure (including OpenAI models)
For models sold by Azure, per Microsoft's region availability documentation:
- Standard and Regional deployments process "prompts and responses in the region associated with your deployment." For example, a UK South deployment stays in the UK, and an Australia East deployment stays in Australia.
- The EU Data Zone processes within any EU member nation.
- The Asia Pacific Data Zone processes within any Asia Pacific nation. This is important to call out: it is NOT an Australia-only guarantee.
For EEA deployments of models sold by Azure, Microsoft's abuse-monitoring human reviewers are located in the EEA (see the region availability documentation).
Most capable in-geography models
The table below is a snapshot of the most capable models available in-geography as of the "last verified" date. Model catalogs change often; treat Microsoft's live tables as authoritative, and use this snapshot only as a point-in-time reference.
Last verified: 2026-07-22. Authoritative live sources: Microsoft's region availability by deployment type and models sold by Azure region availability.
| Geography | Deployment path | Most capable models available |
|---|---|---|
| EU (via EU Data Zone) | EU Data Zone; single-country regional where offered (Sweden Central has the deepest single-country catalog) | GPT-5.6 family, GPT-5.5, GPT-5.4, GPT-5, GPT-4.1, GPT-4o, o3, o4-mini, current embedding models; also Mistral Large 3, Cohere Rerank v4 |
| UK (regional in UK South) | Provisioned-throughput (reserved capacity) for the newest models; standard pay-as-you-go for the rest | GPT-5.4, GPT-5.3-codex, GPT-5.2, GPT-5.1, GPT-5, GPT-4.1, o1 via provisioned throughput; GPT-4o and GPT-4.1-mini on standard pay-as-you-go |
| Australia (regional in Australia East) | Provisioned-throughput (reserved capacity) for the newest models; standard pay-as-you-go for the rest | GPT-5.4, GPT-5.2, GPT-5.1, GPT-5, GPT-4.1, o3, o3-mini via provisioned throughput; GPT-4o and GPT-4.1-mini on standard pay-as-you-go |
One pattern to note: the newest frontier models reach the UK and Australia mainly through provisioned-throughput (reserved capacity) deployments, not pay-as-you-go.
Summary
- OpenAI models used through Azure are hosted by Microsoft; your data is not shared with OpenAI, not used to improve providers' models, and not used to train foundation models without your permission (source).
- Claude models can run either Hosted on Azure (inference on Azure infrastructure) or Hosted on Anthropic infrastructure (inference may leave Azure and your selected region), with Anthropic acting as contractual data processor and not training on customer content under its Commercial Terms (Microsoft, Anthropic).
- Data residency depends on the deployment type. Microsoft's linked tables are the authoritative, live source and should be checked for the current state.