🌐 Overview

Microsoft has introduced the capability for organizations to integrate Anthropic’s Claude models directly into the Microsoft 365 ecosystem. This integration allows IT Administrators to expand the cognitive capabilities of Microsoft 365 Copilot and Copilot Studio by leveraging external Large Language Models (LLMs). Specifically, these models can be utilized within the Researcher agent to enhance information gathering and synthesis tasks.

It is critical for administrators to understand that while these models are accessed through Microsoft interfaces, they are hosted by Anthropic on infrastructure external to the Microsoft-managed environment. This setup facilitates advanced generative AI features such as complex document summarization, multi-source data synthesis, and sophisticated content drafting.

⚙️ Key Technical Details

• 🛡️ Governance and Data Processing: Because Anthropic models operate outside of Microsoft’s boundary, standard Microsoft agreements—including the Product Terms and the Data Processing Addendum (DPA)—do not apply to data processed by Anthropic. Usage is instead governed by Anthropic’s Commercial Terms of Service and Anthropic’s Data Processing Addendum.
• ⚠️ Subprocessor Update & Deprecation: Anthropic has officially onboarded as a Microsoft subprocessor. Consequently, the legacy “opt-in” toggle for using Anthropic under separate commercial terms is being phased out. Effective January 7, 2026, this toggle will be deprecated.
• 🛠️ Functional Capabilities: Once integrated, these models empower the “Researcher” agent to:
  • Condense intricate technical data into digestible summaries.
  • Extract specific answers from provided source materials.
  • Generate and refine drafts for professional documentation.
• 🔑 Prerequisites:
  • Users must be assigned a valid Microsoft 365 Copilot license.
  • The administrator performing the setup must hold the Global administrator role.
• 🚀 Configuration Path (Admin Center):

  To enable the connection, navigate to the Microsoft 365 admin center and follow this path:

  Copilot -> Settings

  Within the Data access page, locate AI providers for other large language models. Under LLM providers for your organization, select Anthropic, review the terms, and select Allow provider.

• 🎛️ Power Platform Controls: For organizations using Copilot Studio, additional granular controls are available within the Microsoft Power Platform admin center (PPAC) to manage external LLM responses.

📊 Impact

🛡️ Compliance and Security Impact: Enabling this connection shifts the data residency and compliance responsibility. Microsoft’s standard audit controls, service level agreements (SLAs), and the Customer Copyright Commitment do not extend to the Anthropic environment. IT Admins must ensure that this external data flow aligns with their organization’s internal risk management policies.

📅 Administrative Latency: When enabling or disabling the Anthropic provider, changes are not instantaneous. It may take several hours for the connection (or disconnection) to propagate across the tenant and for the services to become available or restricted for end-users.

🚫 Service Deactivation: If an organization decides to cease using external LLMs, admins can navigate to Copilot -> Settings in the Microsoft 365 admin center and select Block provider under the Anthropic entry. This will revoke access for all users within the tenant.

Official Source: Read the full article on Microsoft.com