Controlling Data Access for Microsoft Copilot: 6 Steps for Greater Control.

Copilot licenses are distributed quickly. But one of the most important questions before deploying Microsoft Copilot is: What company data can Copilot actually access, and how is that data processed? The basic principle is reassuring at first: Users can generally only view and use information for which they already have the appropriate permissions. Yet this is precisely where a problem arises in many small and medium-sized businesses: All too often, permissions are granted too liberally, or there are legacy permissions that have accumulated over time—and for particularly sensitive data, it may make sense to impose restrictions on Copilot. Furthermore, data is processed differently depending on the model. We’ll explain how to make your Microsoft tenant Copilot-ready.

EBF Mitarbeiter - Copilot steuern

Key Points at a Glance

  • Microsoft Copilot does not override existing access rights: Users can generally only access information for which they already have authorization.
  • This is precisely why historical errors in permissions and overly generous access grants can have a greater impact when using Copilot.
  • Particularly sensitive data should be identified and protected using appropriate permissions, sensitivity labels, DLP policies, and other safeguards.
  • The AI model used and how it is deployed are also relevant: data processing and the providers involved may vary depending on the model.
  • Before rollout, organizations should therefore review permissions, data classes, models, and safeguards, and test the setup with different user roles.

How does Microsoft Copilot generally handle corporate data?

Microsoft Copilot is deployed within the Microsoft 365 architecture. For EU customers, Microsoft’s data protection and data residency commitments—including the EU Data Boundary—generally apply. However, this does not mean that no further review is necessary when using Copilot. While these commitments and the EU Data Boundary generally ensure a good level of data protection, they do not provide complete protection against data transfers to third countries, as there may be exceptions and potential access or transfers outside the EU or EFTA.

In addition, depending on the selected models, there are further specific considerations: For example, some models are operated within Microsoft’s infrastructure, while others may involve external providers as subprocessors. These include, for instance, Anthropic as well as certain models operated by OpenAI. The data protection implications vary depending on the deployment type.

Therefore, there is no simple answer to the question, “Is Copilot compliant with data protection regulations?” Different data protection conditions apply depending on the AI model and model provider, as well as the Copilot feature and tenant configuration. The key to answering this question is: Which Copilot feature uses which model under which configuration for which data?

How to Use Copilot Safely: 6 Steps

  • 1. Review Tenant

    Analyze existing permissions, shares, and protection mechanisms. Identify misconfigurations and overly broad access rights that should be corrected before the Copilot rollout.

  • 2. Identify Sensitive Information

    Determine which data requires special protection and which data may not be processed by Copilot—or may only be processed to a limited extent. Based on this, define the necessary protective measures.

  • 3. Review models and providers

    Determine which AI models and model providers are to be used, and verify the following: Where is their data processed and stored? Is data used for training? Are subprocessors involved, and what contractual data protection provisions apply? Based on this, models may be excluded or restricted to specific use cases as necessary.

  • 4. Technically implement protective measures

    Technically implement the previously defined requirements. This may include, for example, refining permissions and approvals, using Microsoft Purview and sensitivity labels, implementing DLP policies, and configuring settings for the use of specific AI functions and models.

  • 5. Test Different Roles

    Next, use different roles, permissions, and realistic use cases to verify whether the defined protective measures function as intended.

  • 6. Roll Out in a Controlled Manner

    Combine the technical rollout with clear governance rules and user training so that your employees know exactly what they can—and cannot—do.

Insights into practice

EBF-Mitarbeiter

“From our client projects, we know how important it is to take a close look at access permissions that have accumulated over time. Copilot can make it easier to find information that some employees already had broader access to than they needed. That’s why, before rollout, we check whether access rights and safeguards align with actual needs.”

Robert Hamel Solution Architect EBF

Can Copilot be granted access to other enterprise systems?

Copilot delivers particularly significant value when not only Microsoft 365 data is available, but other systems containing corporate knowledge are also integrated. After all, the information employees need for a complete view of customers, processes, or projects isn’t found solely in documents and emails—it’s also hidden in CRM, ERP, HR, or ticketing systems. To answer the question “What’s the status of customer XY?”, it’s ideal to also consider open quotes from the CRM, the current status of tickets, and relevant contract information.

However, one thing is crucial: the authorization structures of the connected systems must be taken into account. Anyone who is not authorized to view information in the source system must not be able to access it via Copilot either. This is not something that can be done on the fly, however, and requires a great deal of expertise.

And that’s exactly why we’ve created EBF Insight—a solution that makes this possible and offers a 360-degree view of relevant information—all in one place, without the hassle of switching between systems. We connect Copilot to additional business applications so that Copilot can access the data stored there in real time. Users’ existing permissions are maintained at all times.

Learn more about EBF Insight

Integration von Unternehmensanwendungen in Copilot

Our conclusion: Microsoft Copilot is not a plug-and-play project.

Using Copilot in compliance with data protection regulations isn’t as easy as it might seem at first. Effective AI governance requires a close look at permissions, tenant settings, model selection, and more. And that’s incredibly important so that your employees can use the tool with confidence. After all, uncertainty is often the reason why true AI adoption fails in small and medium-sized businesses.

Learn more about employee enablement

Contact

Do you have any questions? We’d love to talk with you.

We’re happy to help you implement Copilot in your company in a secure and privacy-compliant manner.

AI Governance Beratung