- Microsoft included Anthropic’s Claude models in M365 Copilot offers choices to users and reduce its dependency on OpenAI.
- Claude in M365 Copilot operates under Microsoft’s data protection safeguards, but data is still transferred outside of Azure.
- Claude models aren’t available in all features and don’t replace the default ChatGPT models in M365 Copilot.
- Anthropic’s Claude models are enabled by default in commercial tenancies starting Jan. 2026 and disabled in the EU and UK.
Microsoft has partnered with Anthropic, adding its Claude models to Microsoft 365 Copilot (M365 Copilot). Claude’s Sonnet and Opus models are popular models that rivaled or surpassed OpenAI’s GPT models. This partnership answers customers’ demand for multi-model options to avoid vendor lock-in and provides users with more tool choices. However, the difference between the two vendors’ latest models isn’t as significant anymore, and choosing models can be confusing for users without adding any real value.
Claude is included with the M365 Copilot license, although early access to AI features, like Anthropic’s Office agents, require enrolling in Microsoft’s Frontier program. Claude is not available in government or sovereign M365 tenancies.
What the Partnership Means
Apart from the circular financial commitments, partnering with Anthropic accomplishes two goals for Microsoft: giving customers a choice of models in M365 Copilot and alleviating its dependency on OpenAI.
Users want choice because some AI models can be better than others at performing certain tasks. For example, Claude is perceived to be better than M365 Copilot’s default model in deep reasoning, automating complex Excel functions, and generating more polished PowerPoint slides from prompts. However, the gap may have narrowed with OpenAI’s GPT 5.2 (rolling out to M365 Copilot), indicating that one model’s advantages can be fleeting, and choice may be a matter of personal preference rather than business value.
More practically, Microsoft is reducing its reliance on OpenAI, its AI partner since 2019. Their relationship has turned rocky in recent months, so diversifying models was expected, especially since GitHub Copilot and Microsoft Foundry also offer choices. Although the OpenAI partnership is not likely to end, the growing pains suggest that more vendors’ models may become available in M365 Copilot.
What About Customer Data?
When the partnership with Anthropic was first announced in Sept. 2025, using Claude models were still subject to Anthropic’s terms and conditions, re-opening customers’ questions about data security and privacy in M365 Copilot. That changed as of Jan. 2026 when Anthropic became what’s called a subprocessor to Microsoft, meaning it’s bound by Microsoft’s Product Terms and Data Processing Addendum (DPA) and doesn’t use customer data for training its models. However, data is transferred from Azure to Anthropic’s servers in AWS or GCP datacenters located primarily in the U.S., unlike OpenAI’s models operating in Azure. Consequently, Anthropic’s designation as a subprocessor to Microsoft may not be enough for some companies subject to the European Union’s General Data Protection Regulation (GDPR) or that have strict internal data protection requirements. However, given the terms of the partnership, it’s reasonable that Microsoft eventually will extend Claude into its European datacenters, but there is no indication of that timeline.
Where to Find Claude
Claude is enabled by default in commercial tenancies starting Jan. 2026, except in the EU, European Free Trade Association (EFTA), and the United Kingdom (UK), where it’s disabled. Government and sovereign tenancies don’t have Anthropic (education tenancies aren’t specified). Anthropic can be enabled or disabled by admins in the M365 admin center’s Copilot settings.
M365 Copilot currently offers Claude Sonnet 4.5 for general queries and Opus 4.5 for structured queries and deep reasoning. Haiku 4.5 is not included. Users can choose Claude in the following places, although it’s not always clear which model is used for some features:
- Researcher Agent: Utilizes the deep reasoning of Claude Opus 4.1 (it may be 4.5 now but it’s not documented)
- Agent Mode in Excel: Claude Opus 4.5 is available in Agent Mode in Excel for the web while Agent Mode with Claude is rolling out to Excel for Windows
- Excel, PowerPoint, and Word agents: These Anthropic Office agents are available through the Frontier program (Microsoft does not offer corresponding agents)
- Copilot Studio: Claude Sonnet 4.5 and Opus 4.5 are available for custom Copilot agents, but both are labeled “experimental”
- Copilot with Claude: This agent-like feature is used in the M365 Copilot chat interface; however, it’s rolling out and is not available everywhere yet.
For more details, see the sidebar: Where to Find Claude in M365 Copilot.
Limitations and Considerations
Although it’s now available, Claude can’t fully replace the default OpenAI model in M365 Copilot. Instead, Claude will likely expand into more features such as Copilot Chat and Agent Mode for Word and PowerPoint where users can select it.
There are no management controls yet, so after Claude is enabled in a tenancy, everyone with an M365 Copilot license (and enrolled in the Frontier program for preview features) will have access. As the offering matures, customers should expect to manage access to Claude using security groups, although whether it will be at the provider or feature level is questionable.
Although Claude adds choices for M365 Copilot users, it can also create confusion regarding when to use which model. While it may end up being a personal choice, organizations should provide training and guidance to their employees and update relevant governance policies.
Claude is included with the M365 Copilot license, raising questions about the sustainability of Microsoft’s fixed-fee pricing. While not a direct comparison, Claude Sonnet costs roughly 1.5 to 2 times more than GPT 5.1, while Opus is 7 to 10 times pricier, based on Microsoft Foundry rates per million tokens. If Claude sees widespread use, M365 Copilot’s licensing may shift to a tiered or base-plus model.
Licensing & Availability
Anthropic models are included with the M365 Copilot license for no additional cost. M365 Copilot is an add-on license for M365 subscriptions.
Enrollment in the Frontier program is required for preview features like Anthropic’s Office agents.
Directions Recommends
Disable Anthropic immediately if customers have any questions. Microsoft is enabling Anthropic in commercial M365 tenancies outside of the EU Data Boundary. Despite Microsoft providing data protection, data is still transferred to Anthropic. If companies have concerns, they should disable Anthropic until those questions are answered.
Verify that subprocessor status meets regulatory and data residency requirements. Although the subprocessor status provides some protection for customers, data is transferred out of Azure to Anthropic’s datacenters, mainly located in the United States. This may not meet more stringent regulatory or internal data processing requirements of some customers.
Avoid user confusion by providing education and training. Users may not know the differences between M365 Copilot models, leading to wasted time testing options and increased frustration. Training and guidance help, but the lack of management controls complicates an ordered rollout of Claude.
Resources
Including Anthropic’s models in M365 Copilot was introduced in the Directions blog post “Microsoft Begins Rolling Out More Model and Agent Options for M365 Copilot and Beyond.”
Claude models in Microsoft Foundry is discussed in the Directions report “Foundry Gets New Name, Anthropic Models.”
Anthropic as a subprocessor for Microsoft is described in “Anthropic as a subprocessor for Microsoft Online Services” (Microsoft).
The partnership with Anthropic is announced in “Microsoft, NVIDIA and Anthropic announce strategic partnerships” (Microsoft).