Microsoft just gave every Microsoft 365 Copilot admin a real deadline. The Copilot OpenAI Subprocessor setting rolled into the admin center on July 9, 2026, starts disabled, and switches itself on for eligible tenants on July 24, 2026 if nobody touches it (MC1422074). That is fourteen days to make a deliberate choice instead of getting a default one.
This is not a routine feature announcement. Microsoft has onboarded OpenAI as a subprocessor so Microsoft 365 Copilot can run OpenAI-operated models, starting with GPT-5.6, alongside the models it already runs on Azure OpenAI. The Copilot OpenAI Subprocessor setting is the toggle that decides whether your tenant uses that new option or not. This post covers what actually changes, why the deadline matters more than the feature, how the rollout works, and the Paul-Take on what to do this week.
What the Copilot OpenAI Subprocessor Setting Actually Does
The Copilot OpenAI Subprocessor setting lives in the Microsoft 365 admin center and controls whether Microsoft 365 Copilot, Copilot in Microsoft 365 apps, and Microsoft Copilot Studio are allowed to call OpenAI-operated models directly, rather than only models running on Microsoft’s own Azure OpenAI infrastructure.
What changes if you enable it
- Access to OpenAI’s newer model family, beginning with GPT-5.6, inside Microsoft 365 Copilot
- Faster access to future AI model innovation as OpenAI ships new releases
- Prompts and responses still governed by the Microsoft Product Terms and Data Protection Addendum, with the enterprise data protection commitments you already rely on
What stays the same either way
- Models running on Microsoft Azure OpenAI are unaffected, this setting does not touch them
- No change to existing Copilot licensing, this is a configuration toggle, not a new SKU
- Admins can flip the setting back and forth at any time in the admin center
Why the Copilot OpenAI Subprocessor Deadline Matters
Most Message Center items give you months to plan. This one gives you two weeks, and the default outcome is enablement, not the safer option. If your organization has any policy against using non-Microsoft-operated AI infrastructure, or if legal and compliance have not signed off on an additional subprocessor, July 24 is the date that decision gets made for you unless you act first.
The subprocessor question is not new, just louder
Anyone who tracked the earlier Subprocessor Disclosure notice already knows Microsoft periodically adds vendors to the list of parties that can touch tenant data under specific conditions. What is different with the Copilot OpenAI Subprocessor setting is that it is Copilot-facing and tied to a hard auto-enable date, which is exactly the kind of notice our Subprocessor Disclosure post warned admins not to skim past.
Model choice inside Copilot is becoming a real conversation
This also is not the first time Copilot admins have had to actively decide which model runs behind their AI experiences. When Copilot Cowork introduced an admin-controlled option to opt in to a non-Microsoft-operated model, the same pattern showed up, real governance choice, not a checkbox nobody reads. The Copilot OpenAI Subprocessor setting extends that pattern to mainstream Microsoft 365 Copilot itself.
How the Copilot OpenAI Subprocessor Rollout Works
Rollout timeline
| Stage | Window |
|---|---|
| Setting appears in admin center (Worldwide) | Began July 9, 2026 |
| Default state | Disabled |
| Auto-enable date for eligible tenants | July 24, 2026 |
| Admin action required | Yes, before July 24, 2026 if you want to keep it off |
What to do before July 24
- Find the OpenAI-operated models setting in the Microsoft 365 admin center now, do not wait for a reminder
- Loop in whoever owns AI governance or data protection policy at your organization, this is their call as much as IT’s
- If you want OpenAI-operated models available, review the linked Microsoft Learn documentation and enable deliberately
- If you do not want them, select ‘No users’ before July 24, 2026 so the auto-enable never triggers
- Document the decision either way, this is the kind of setting that gets asked about in a security review later
Admin Tips
- Treat the default as ‘on’ for planning purposes, because that is what happens if nothing is done
- Selecting ‘No users’ does not block Copilot itself, it only blocks the OpenAI-operated model path, Azure OpenAI-hosted models keep working
- Brief anyone running Copilot Studio agents separately, this setting also affects Copilot Studio, not just the Copilot chat experience
License requirements
The Copilot OpenAI Subprocessor setting requires no new license and no additional per-user cost. It is a configuration option available to admins already managing Microsoft 365 Copilot and Microsoft Copilot Studio. No GCC High or DoD considerations have been published for this rollout at this time.
The Paul-Take on the Copilot OpenAI Subprocessor Setting
Honest opinion. The feature itself, access to GPT-5.6 and future OpenAI models, is not the story here. The story is that Microsoft picked an opt-out model with a hard date, and a lot of tenants will sail past July 24 without anyone having made a real decision. That is exactly backwards for a change involving a new subprocessor touching AI prompts and responses.
If your organization has any formal stance on approved AI vendors, data residency, or subprocessor lists, this needs a decision this week, not a shrug. The upside of OpenAI-operated models is real, quicker access to model improvements is genuinely useful for Copilot power users. But ‘genuinely useful’ and ‘defaulted on without review’ should never be the same sentence in a compliance conversation. Similar to how Claude Fable 5 forced Copilot Cowork admins to actively choose a non-Microsoft model rather than inherit one, this setting deserves the same explicit choice, just with a much shorter runway.
My advice, do not let this become the Message Center item you find in your archive three months from now wondering when it got turned on. Check the setting today, get the right person to sign off, and pick a side before July 24.
This should be rolling out in admin center configuration from July 9, 2026 and auto-enabling from July 24, 2026 according to Microsoft.
Quick checklist for the Copilot OpenAI Subprocessor deadline
- Located the OpenAI-operated models setting in the Microsoft 365 admin center
- Looped in AI governance or compliance owner before July 24, 2026
- Decided to enable or select ‘No users’ with a documented reason
- Briefed teams managing Copilot Studio agents separately
- Confirmed Azure OpenAI-hosted model usage is unaffected either way
Related Resources
The Copilot OpenAI Subprocessor setting sits inside a bigger AI governance picture. Pair this post with the rest:
Microsoft official references:
- OpenAI as a subprocessor in Microsoft Online Services
- Enterprise data protection in Microsoft 365 Copilot
MVP Reference List
- Message Center: MC1422074
- Action required by: July 24, 2026
- Microsoft Learn, OpenAI subprocessor: openai-subprocessor
- Microsoft Learn, enterprise data protection: enterprise-data-protection