Microsoft is rolling out Anthropic's Claude models as a built-in option inside Microsoft 365 Copilot, Copilot Studio, and several Microsoft 365 apps โ with full availability expected by the end of March 2026. For federal agencies on commercial M365 tenants, this introduces new model choices and new compliance questions. For GCC, GCC High, and DoD tenants, the answer is simpler: Claude isn't coming yet.
What Changed
Starting January 7, 2026, Anthropic formally became a subprocessor for Microsoft Online Services. This is a structural change: Anthropic is no longer a separate vendor with a separate data processing agreement. Microsoft now governs Claude's use through its own Product Terms and Data Protection Addendum (DPA), and the Microsoft Customer Copyright Commitment extends to Claude outputs.
The practical effect: users in eligible tenants will see Claude as a model option in M365 Copilot (web, desktop, mobile), in Copilot Studio agent creation, in Excel's Agent Mode, and in the Researcher experience. Microsoft is enabling it by default for most commercial cloud customers outside the EU, EFTA, and UK โ which have Claude disabled by default due to data boundary commitments.
The Government Cloud Exception
Federal agencies on GCC, GCC High, or DoD tenants won't see a Claude toggle at all. Per Microsoft's documentation:
"Anthropic models aren't available in government clouds (GCC, GCC High, DoD) as there's no FedRAMP certification in place yet."
This is straightforward. FedRAMP Authorization is the threshold for federal cloud deployments, and Anthropic's models have not cleared that bar. Until they do, Claude is not available in government tenants โ not as an opt-in, not as a preview.
This is the right call. The absence of a toggle is a feature, not a gap: it means the boundary is being enforced rather than left to each agency to sort out.
Why This Matters for Federal IT Leaders
The Anthropic rollout surfaces a dynamic that federal IT leaders will encounter repeatedly over the next few years: commercial AI capabilities will arrive in M365 faster than they can be validated for government use. The gap isn't Microsoft's fault โ it's the natural friction between a fast-moving commercial AI ecosystem and the deliberate, risk-managed certification process that federal data environments require.
What to watch:
- FedRAMP trajectory for Anthropic: There is no public timeline for Anthropic pursuing FedRAMP authorization. Until that changes, Claude remains unavailable for government cloud.
- Commercial tenant vs. GCC tenant decisions: Agencies with both commercial and government tenants need clear policies on which workloads live where. Claude's availability in commercial tenants while absent from GCC is exactly the kind of asymmetry that causes drift.
- Model proliferation governance: Even within GCC tenants, new models will arrive over time. Agencies should be building the governance muscle now โ acceptable use policies, model-by-model review processes, DLP configuration โ rather than reacting to each new rollout.
The Compliance Posture Clarification
For any federal agency that has been considering Anthropic models via Microsoft, there is one important administrative change: the previous opt-in path using Anthropic's own commercial terms and DPA is being deprecated. Going forward, if and when Claude becomes available in government environments, it will be governed exclusively under Microsoft's enterprise framework โ not Anthropic's separate terms.
That's a meaningful improvement in compliance posture for agencies that were navigating dual agreements. It simplifies the vendor relationship and puts Microsoft squarely on the hook for Claude's behavior within M365 services.
Bottom Line
GCC, GCC High, and DoD tenants: nothing changes for you today. Claude is not available in your environment and no action is required. The right question to monitor isn't "when can we use Claude?" โ it's "what is Anthropic's FedRAMP roadmap, and does that timeline align with our AI strategy?"
For agencies on commercial M365 tenants, the governance questions are real. Review your admin center settings, verify your DLP policies are scoped correctly, and assess whether multi-model Copilot use requires updated acceptable use guidance for your users.
Source: Anthropic as a subprocessor for Microsoft Online Services, Microsoft Learn, updated March 4, 2026.