Microsoft has started to broaden their AI horizons by adding their first (not Open AI) model into Copilot.
Microsoft are integrating Anthropic’s Claude models into Microsoft 365 Copilot which marks a significant pivot from their exclusive OpenAI-centric approach. Microsoft are also working on their models which we already see on Copilot Plus PCs which will at some point make their way to Copilot.
This move is more than just a new menu option or toggle, it is part of their strategic play to diversify AI capabilities and reduce dependency on a single vendor.
Claude Opus and Sonnet in Copilot.
Claude Opus 4.1 and Sonnet 4 are now available to commercial Frontier Copilot users, (Corporate early adoptors) offering, for the first time, alternatives to Open AI’s GPT models for agents in Copilot Studio and also for their Researcher Agent.
It’s worth noting that enabling access does require admin approval. See later.
In the formal annoucement, Microsoft said that Anthropic’s models will unlock “more powerful experiences” for users.
Claude is not new to Microsoft however, with it already embedded in Visual Studio and Azure AI alongside Google’s AI and Elon Musk’s Grok. This is, however the first time Copilot launch that we have seen non OpenAI models powering Copilot.
Why This Matters
Microsoft’s shift to leveraging different models reflects a broader trend. Microsoft’s message here is that Copilot is no longer about a single model or even vendor, bit more about orchestration, choice, and adaptability.
Different models have different areas of excellence and this sets the foundations for Microsoft to give flexibilityto tailor and tune AI experiences to specific business needs using the most appropriate model for the task.
It does, however, raise questions around governance, model performance, and cost. With multiple models in play, we don’t really know how the future of pricing will work if multi model is the future for Microsoft 365 Copilot.
Data Sovereignty and Multi-Model Concerns?
One question I’m already seeing is around Microsoft’s boundary of trust and responsibility, something Microsoft boast around with their Microsoft 365 portfolio.
While the flexibility of multi-model AI is compelling, the question is does it introduce new considerations around data residency and compliance when multi models are being used?
To address that, Microsoft has confirmed that these Claude models run within its Azure AI infrastructure, but states that are not Microsoft-owned. This means that when users “opt to” use Claude, their prompts and responses may be processed by Anthropic’s models hosted within Microsoft’s environment.
This means that when organisations choose to use Anthropic models, they are using these under Anthropic’s Commercial Terms of Service, not the consumer user terms.
For regulated industries or organisations with strict data governance policies, this is likely to raises a few red flags or at least questions that Microsoft will need to be able to answer.
- Data Boundary Clarity: Is the data staying within Microsoft’s compliance boundary, or is it crossing into Anthropic’s operational domain? If so what does this mean for data compliance and security?
- Model-Specific Logging: Are logs and telemetry handled differently across models? Can organisations audit usage per model? How is encrypted data handled?
- Privacy and Consent: Are users aware when their data is being processed by a non-Microsoft model? Is consent granular enough? Will users understand even if Microsoft tell them?
Again, Microsoft has stated that Claude models are “fully integrated” into the Microsoft 365 compliance framework, but organisations will still want to (and should) validate this against their own risk posture – especially where sensitive or regulated data is involved.
Enabling Claude models in Copilot.
To enable the models, your Microsoft 365 Admin needs to head over to the Microsoft 365 Admin Centre and enable access to the other models. Instructions for this are shown in the link below.
https://learn.microsoft.com/en-us/copilot/microsoft-365/connect-to-ai-models?s=09
Thoughts.
This is a smart move I think. Microsoft is playing the long game — moving their eggs out of one basket and looking a different models that made most economic and performance sense and brining more choice to agent builders.
For those of us partners like us at Cisilion, advising clients on AI adoption, this reinforces the need to think modularly. When building agents, don’t just pick a model – pick a framework that allows you to evolve. Microsoft’s Copilot is becoming that framework and that should be good for business.
I do expect this is just the start. We know Microsoft’s relationship with OpenAI is “less properpous” that it once was. As such I do expect more models, more integrations, and more choice and I do think we will see Microsoft’s own models making their way to Copilot soon.
But with choice comes complexity. We need to ensure that governance, transparency, and user education keep pace with innovation. Again partners will need to help customers navigate this.
What do you think. Is this a good move for Microsoft and their customers?