Choosing the right AI model in Microsoft 365: flexibility, control, and confidence

AI innovation is moving fast, and no single model is perfect for every task. Some models excel at deep research and structured reasoning, while others are optimized for speed, summarization, or creative drafting. That’s why Microsoft 365 is evolving to support AI model choice, giving organizations more flexibility while maintaining enterprise-grade governance.

Whether you’re an IT admin setting policy, a maker building agents, or an end user working in Copilot, here’s how AI model choice works across Microsoft 365 today.

A multi-model approach, built for enterprise needs

Microsoft 365 Copilot uses a multi-model architecture. Microsoft-hosted models remain the default for Copilot experiences, but organizations can also choose independent AI models or AI subprocessors for specific scenarios.

This approach is intentional. Different models bring different strengths, and enterprises need the ability to match the right model to the right task without compromising security, compliance, or control.

AI subprocessors: What it means for admins

Some AI models are onboarded as Microsoft subprocessors for Microsoft Online Services. For example, Anthropic models can now be used under Microsoft’s contractual framework when enabled by an admin.

• Microsoft Product Terms and the Microsoft Data Protection Addendum (DPA) apply.
• Admins retain centralized control over whether the model is available.
• Usage aligns with Microsoft’s enterprise governance commitments and Enterprise Data Protection.
• Subprocessors are contractually required to meet Microsoft security, privacy, and compliance standards.

Learn more: Overview of AI Subprocessors in Microsoft 365 Copilot 

Independent AI models: opt-in flexibility

Microsoft 365 also supports connecting to independent AI models that are hosted outside Microsoft-managed environments. Today, this includes models from providers such as Anthropic and xAI.

• Data may be processed outside Microsoft’s managed infrastructure.
• Provider-specific terms and data handling policies apply.
• Admin approval is required before models become available to users or makers.
• Organizations should review regional and data residency considerations before enabling these models.

Learn more: Understanding AI functionality and models in Microsoft Online Services

Admin control in the Microsoft 365 admin center or Power Platform admin center

For organizations building custom agents or copilots, model choice is governed centrally.

• Allow or block external language models.
• Control which models are available for generative responses.
• Balance innovation with organizational risk tolerance.

Model choice for makers: Copilot Studio

In Microsoft Copilot Studio, makers can select the AI model that powers an agent’s reasoning or responses when allowed by admin policy.

• Choose a primary AI model for an agent.
• Switch models to compare behavior and performance.
• Use external models where enabled or fall back to Microsoft-hosted defaults.
• Preview and experimental models are clearly labeled and not recommended for production use.

End-user experiences: Choosing the right model for the task

The Researcher agent in Microsoft 365 Copilot helps you gather, analyze, and summarize information from both the web and your work content.

With model choice enabled, Researcher can use multiple AI models, including GPT models from OpenAI and Claude models from Anthropic, to generate research reports using different reasoning approaches.

In Auto mode, Researcher generates a report using GPT and applies a second reasoning pass with Claude to strengthen structure and completeness, prioritize reputable sources, and ensure key statements are grounded in citations.

For more advanced workflows, Model Council mode runs the same question through multiple deep-reasoning research agents in parallel and highlights where outputs align or diverge so users can compare perspectives and combine insights with greater confidence.

Learn more:

• Data, Privacy, and Security for Microsoft 365 Copilot
• Use model choice in the Researcher agent
• Use Claude with Researcher in Microsoft 365 Copilot

Choosing responsibly: A shared responsibility model

• Microsoft provides secure defaults, transparency, and enterprise controls.
• Admins decide which models are allowed and where.
• Makers and users select models intentionally based on the task at hand.
• Existing Microsoft 365 security, compliance, and permission controls continue to apply when using Copilot experiences.

Privacy, security, and compliance considerations

Microsoft 365 Copilot is compliant with existing Microsoft 365 privacy, security, and compliance commitments. Prompts, responses, and Microsoft Graph grounding data are not used to train foundation large language models used by Microsoft 365 Copilot. Organizations should also review regional data residency guidance, including EU Data Boundary considerations, when enabling AI subprocessors.

Learn more:

• Enterprise data protection in Microsoft 365 Copilot and Microsoft 365 Copilot Chat
• Licensing Documents 

Feedback

AI model choice in Microsoft 365 is evolving, and your input helps shape what comes next. Whether you’re an IT admin thinking about governance, a maker experimenting with agents, or an end user trying different Copilot experiences, we want to hear what’s working and what isn’t.

Share your thoughts, questions, or scenarios in the comments below. Your feedback helps inform future updates, documentation, and product decisions as Microsoft continues to expand AI model choice across Microsoft 365.

Learn about the Microsoft 365 Insider program and sign up for the Microsoft 365 Insider newsletter to get the latest information about Insider features in your inbox once a month!