OpenAI becomes a subprocessor in Microsoft 365: what you need to check before 24 July

Starting July 24, 2026, Microsoft will change an important setting within Microsoft 365 Copilot. Models managed directly by OpenAI will then become available to users by default, unless your organization makes a different choice in advance.

That doesn't automatically mean you should turn off these models. It does mean that as an organization you have to consciously determine whether you want to use them, for which employees and under what conditions.

This raises questions about privacy, data processing, security and the functionalities that your employees can use. In this blog, we explain what will change, what choices you have and what you need to check before 24 July.

What will change on 24 July?

Microsoft 365 Copilot has been using OpenAI technology for a long time. These models are offered by Microsoft and run within the infrastructure that Microsoft manages. Recently, Microsoft also offers models that are managed directly by OpenAI. OpenAI then processes the data needed to make the model work on behalf of Microsoft.

At the moment, these models are turned off by default. Starting July 24, they'll be available to all users within eligible Microsoft 365 commercial environments, unless an administrator modifies the setting in advance. Microsoft indicates that this gives organizations more choice between different AI models. This also allows new models to become available more quickly within Microsoft 365 Copilot.

For organizations, this means that choosing and assessing AI models is becoming an increasingly important part of AI governance. It is no longer just about whether employees are allowed to use AI, but also which models and suppliers are allowed to do so.

What role will OpenAI have?

When you use models managed by OpenAI, OpenAI processes certain data on Microsoft's behalf. In legal terms, OpenAI thus becomes a subprocessor.

The agreements that your organization has with Microsoft will continue to apply. Microsoft remains the contractual point of contact and OpenAI may only use the data to provide the agreed service.

Yet something important is changing. An additional party will be added to process data and the processing will partly take place on systems managed by OpenAI. That doesn't have to be a problem, but it is something that you as an organization have to consciously look at. Does this working method fit within your agreements on privacy, information security and compliance? And is it clear who makes a decision about this?

What does this mean for your organization?

When choosing whether or not to make these models available, four questions are especially important.

Where is data processed?

Microsoft indicates that the processing for European organizations remains within the European data border in most situations.

There is a separate setting that allows processing to temporarily take place outside Europe during busy times. This setting is separate from the choice to make OpenAI models available. You should therefore check both settings separately.

Does this fit within your policy?

Some organizations have strict requirements for the countries or regions in which data may be processed. This can arise from legislation, contractual agreements, internal policies or customer requirements.

For organizations that require data to be processed within one specific country, there is an important point of attention: the models that OpenAI manages are not currently covered by this commitment. Therefore, check whether the new working method is in line with the agreements that apply within your organization.

Who gets access?

You don't have to make the models available to the entire organization right away. You can choose from three options:

1. Access for all users.
2. Access for a limited group of employees.
3. No entry.

A limited group can be a good intermediate step. This way you can first gain experience, assess the use and check whether the agreements made work in practice.

Which functionalities do you want to use?

The choice does not only have consequences for privacy and data processing. Some new models and functionalities are only available when OpenAI is allowed as a subprocessor.

A current example is GPT-5.6. This model is available within Word, Excel, PowerPoint, Copilot Chat and Copilot Cowork, among others. To use it, access to models that OpenAI manages must be enabled. Therefore, include both the risks and the desired functionalities in your decision-making.

Can processing take place outside Europe?

In addition to the setting for OpenAI, you should also look at flex routing. This is a separate institution for organisations in the EU and EFTA. If flex routing is enabled, Copilot can temporarily process it outside the European data border during peak loads, for example in the United States, Canada or Australia. The data remains encrypted during transmission and at rest. In principle, the storage of data remains within the European data limit. Limited pseudonymized data may be stored outside this boundary for security and operational purposes.

If flex routing is disabled, the processing remains within the European data limit even during peak loads. For environments created after March 25, 2026, flex routing is enabled by default according to Microsoft. For older environments, Microsoft refers to the communication in the Message Center. Therefore, do not make assumptions based on the age of your environment. Check which choice is actually set.

What choices can you make as an organization?

Do you first want to determine internally whether and for whom you will make the models available? Then you can adjust the setting before July 24. You can disable the models for all users, make them immediately available to the entire organization or only give access to specific employees or groups.

That choice applies to both Microsoft 365 Copilot and Copilot Studio. This makes it important to look not only at individual Copilot users, but also at employees who build agents or solutions themselves.

For Microsoft 365 admins

You need the AI Administrator or Global Administrator role to customize the setting. In the Microsoft 365 admin center, go to:

Copilot > Settings > View all > AI providers operating as Microsoft subprocessors

Then, select OpenAI and choose which users get access.

In addition, check the setting:

Copilot > Settings > View all > Flex routing during peak load periods

This way you not only know whether OpenAI models are available, but also where processing may take place during peak loads.

What do you need to arrange before 24 July?

You don't have to make the same choice for all employees right away. You can disable OpenAI, make it available to the entire organization or start with a limited group first.

It is especially important that the attitude does not change unnoticed. Therefore, before July 24, discuss:

1. Whether your organization wants to use these models.
2. Which employees get access.
3. Whether the processing fits within your privacy, security and compliance policy.
4. Which setting is active for processing during peak load.
5. Who records the decision and assesses future changes.

For many organizations, using these models can be an acceptable choice. The most important step is that the decision is made consciously, is well substantiated and fits within the agreements on responsible AI use.

New models and vendors are becoming part of Microsoft 365 at an ever-increasing rate. A fixed process for assessing these types of changes helps prevent a technical adjustment from automatically becoming an organization-wide choice.

FAQs on OpenAI as a Subprocessor

Will OpenAI be automatically enabled on July 24?

Starting July 24, 2026, OpenAI operated models will become available by default to all users at eligible commercial customers. If you don't want this, or if you want to restrict access to a certain group, an AI Administrator or Global Administrator must adjust the setting in the Microsoft 365 admin center.

Isn't Microsoft 365 Copilot already using OpenAI models?

Yes. Microsoft 365 Copilot already uses OpenAI models that Microsoft manages itself through Azure OpenAI. The new setting specifically concerns models managed by OpenAI where OpenAI acts as a subprocessor of Microsoft.

What is a subprocessor?

A subprocessor is a third party that processes data on behalf of Microsoft in order to provide a service. Microsoft will continue to be the contractual point of contact, and the Microsoft Product Terms and Data Protection Addendum will continue to apply.

Is our data used to train OpenAI models?

According to Microsoft, prompts, responses, and data from Microsoft Graph are not used to train the underlying foundation models. Enterprise Data Protection and the agreements with Microsoft will also continue to apply to OpenAI as a subprocessor.

Is our data leaving the European Union?

OpenAI operated models fall within the EU Data Boundary, according to Microsoft. When flex routing is enabled, LLM processing during peak loads can temporarily take place outside the EU Data Boundary, for example in the United States, Canada or Australia.

What is the difference between the OpenAI setting and flex routing?

The OpenAI setting determines whether users are granted access to models managed by OpenAI. Flex routing determines whether LLM processing may take place outside the EU Data Boundary during peak loads. For a full review, you should check both settings.

Can we only enable OpenAI for a pilot group?

Yes. You can restrict access to specific users or Microsoft Entra ID security groups. This allows you to first gain experience with a limited group and evaluate the consequences for use, privacy and governance.

What happens if we disable OpenAI?

Users will then not be able to use models managed by OpenAI. As a result, certain models and functionalities, including GPT-5.6, may not be available. You can adjust the setting again at a later time.

Who can change the setting?

The setting can be customized by an administrator with the AI Administrator or Global Administrator role.

Does this change also apply to government and sovereign cloud environments?

OpenAI operated models are currently not available within GCC, GCC High, DoD and sovereign clouds.

Should we disable OpenAI as a subprocessor?

That depends on your organization's privacy, security, compliance, and functionality requirements. There is no choice that is automatically the right one for every organization. Assess what data processing is acceptable, what functionalities are needed and for which users you want to make it available.

Do you know which AI models are active within your organization?

New models and vendors are becoming part of Microsoft 365 at an ever-increasing rate. This also increases the need to make clear agreements about access, data processing, ownership and supervision.