Exercise - Enable a third-party AI provider

  • 10 minutes

You need to make one deliberate provider decision for Relecloud, reasoning through the data-exposure trade before regulated Woodgrove Bank work runs through Copilot.

Important

Prerequisites for this exercise:

  • A Microsoft 365 tenant with Microsoft 365 Copilot provisioned and Global Administrator access—connecting a third-party AI provider requires this higher role, not just AI Administrator.

If your lab tenant restricts enabling a live third-party provider, complete Task 1 up to the review step and use Task 2's reasoning exercise regardless.

Task 1: Enable a third-party AI provider

  1. Sign in to the Microsoft 365 admin center at admin.microsoft.com with your Global Administrator account.
  2. Go to Copilot > Settings > View all > AI providers for other large language models.
  3. Review the current provider catalog under this control—Mistral and xAI—each shown with an independent enable/disable toggle. (Anthropic and OpenAI live under a separate AI providers operating as Microsoft subprocessors control and need only AI Administrator.)
  4. Choose one provider appropriate for a specific Relecloud use case (for example, enabling a provider known for a particular reasoning or coding strength for a pilot team), turn it on, and save.
  5. Confirm: this action required Global Administrator, not AI Administrator—reflecting that this is a tenant-wide data-boundary decision, a deliberately higher bar than the rest of the settings catalog you configured in Unit 3.

Task 2: Articulate the data subprocessor trade-off

  1. Write a short justification (two to three sentences) for the pilot team that will use this provider, covering: which provider you enabled, what capability it adds that Microsoft's own models don't (or don't do as well) for this use case, and what data-exposure trade-off the tenant is accepting by routing some prompts to a non-Microsoft subprocessor.
  2. Reflect: enabling the provider was the easy half of this task. The harder discipline—and the one Woodgrove's reviewers will actually ask about—is being able to explain in writing exactly which data now flows to a new subprocessor, and why that trade was worth making for this specific team's need.

You've now enabled a third-party AI provider using the correct elevated role, and articulated the data subprocessor trade-off that decision accepts—exactly the deliberate, documented choice Relecloud needs before regulated work runs anywhere near it.