Recently, a notification from the Bundeshaus (the seat of the Swiss Parliament) caused a stir: the Microsoft Copilot icon had suddenly appeared on the screens of parliamentarians. The reaction was one of uncertainty and criticism, fueled by the narrative that a US corporation was unilaterally deciding on the technology used at the heart of Swiss democracy. This concern was amplified when parliamentary services announced that deactivating the feature was "only partially possible." [1]
While the concern for digital sovereignty is both valid and crucial, a closer technical examination reveals a different story. The situation is less dramatic than it first appeared, but it highlights a much more fundamental challenge that many organizations face in the cloud era.
1. Not Every Copilot Reads Your Emails
The most significant misunderstanding must be addressed first. When a colorful new icon appears in the taskbar, it is most likely not the fully licensed "Microsoft Copilot" (until recently branded Microsoft 365 Copilot) that reaches internal emails, SharePoint documents, or Teams chats on its own.
Instead, this version is "Microsoft Copilot Chat" (until recently branded Microsoft 365 Copilot Chat, and before that "Copilot with Commercial Data Protection" or Bing Chat Enterprise). Here are the verified facts about this service, according to official Microsoft documentation:
- Not Grounded in Internal Data by Default: Copilot Chat is grounded in web data only and has no access to Microsoft Graph. It is not, however, blind to organizational content. Microsoft documents four ways in which content reaches it: the user pastes or uploads it with the prompt, the user works with open content (the Word, Excel and PowerPoint agents, the Copilot mobile app, or an intranet page in Edge), the user runs Copilot Chat in Outlook, where it can reach their mail, calendar, meetings, chats and limited OneDrive and SharePoint file content through lexical indexing, or the organization enables a pay-as-you-go agent that is grounded in tenant data. [5]
- Contractual Data Protection: Microsoft guarantees that user prompts and responses are not used to train its foundation AI models. This is a core component of its Enterprise Data Protection (EDP) commitment. [3]
- A Secure Alternative: It provides a secure, enterprise-grade environment for employees who might otherwise turn to unprotected public AI tools for work-related queries, thereby creating a real security risk.
Microsoft now separates three tiers rather than two: Copilot Chat (Basic), Microsoft 365 Copilot (Basic), and Microsoft 365 Copilot (Premium). Only the Premium tier reaches organizational data automatically, through Microsoft Graph and Work IQ. [6]
2. "Turning It Off Isn't Possible" – Yes, It Is.
The statement that deactivation is not possible is, from a technical standpoint, incorrect. While it is true that Microsoft does not provide a simple "off switch" in a user-facing menu, administrators have complete control.
According to Microsoft's own documentation, Copilot can be managed and disabled organization-wide through the Microsoft 365 admin center (Integrated Apps), Teams app permission policies, and the Microsoft Edge sidebar policy. [2] The fact that this was communicated as "not possible" in the Swiss Parliament points less to a mandate from Microsoft and more toward the complexities of internal IT processes or a potential knowledge gap in managing these new, rapidly evolving service plans.
One caveat has been added since: as of 28 January 2026, Copilot Chat can no longer be unpinned from the navigation of the Microsoft Copilot app for any tenant worldwide. Unpinning was never the control that mattered — blocking the Copilot app through Integrated Apps still removes access tenant-wide — but the "just unpin it" answer is now wrong. [2]
This assessment is supported by the NZZ report, which noted that the Swiss bank UBS successfully implemented a controlled rollout of Copilot. UBS initially blocked the feature for all employees, then gradually enabled it—first for staff without access to customer data, and later for all employees with a filter to detect sensitive information. [1]
3. The Real Issue: Governance is the Customer's Responsibility
The incident in the Swiss Parliament is symptomatic of a challenge facing countless organizations today. Cloud providers like Microsoft operate on an "evergreen" model, rolling out new features at a rapid pace. That is their business model.
However, the responsibility for governing how these tools are deployed and used lies squarely with the customer—the company or the public authority. Digital sovereignty is not achieved by blocking technology, but by actively steering it.
This is the core of the Shared Responsibility Model, a framework clearly defined by Microsoft: [4]
| Responsibility | Microsoft | Customer |
|---|---|---|
| Cloud Infrastructure & Platform Security | ✔ | |
| Data Governance & Classification | ✔ | |
| User Access Management | ✔ | |
| Policy Definition & Enforcement | ✔ |
One responsibility does not appear in Microsoft's matrix at all, and it belongs in the same conversation: user training and enablement. Nobody else is going to do it for you.
Conclusion: Technology Requires Leadership, Not Reaction
Confronting parliamentarians with a new AI tool without warning or training was the actual failure in this scenario. The problem was not the technology itself, but the absence of a proactive change management and governance strategy.
Uncertainty is a byproduct of ignorance. Any organization introducing AI—whether intentionally or as a result of an automatic update—must ensure three things:
- Technical Clarity: Understand what the tool can and cannot do.
- Robust Governance: Establish clear guidelines that go beyond basic data protection laws.
- User Enablement: Provide training so that users are not just passengers but confident pilots of the new technology.
Zero Trust. Zero Drama. Zero Bullshit.
Need support with your AI governance strategy? It is perfectly normal for internal IT departments to feel overwhelmed by the sheer speed of AI development. I can help you separate fact from fiction, configure your environment securely, and empower your employees for the future. So you can decide how AI works for you—not the other way around.
Feel free to contact me: easym365.de
References
- NZZ, "Vertrauliche Dokumente als Trainingsdaten für amerikanische KI? Die Computer im Parlament wurden von Microsoft eigenmächtig umgerüstet," January 4, 2026. https://www.nzz.ch/schweiz/vertrauliche-dokumente-als-trainingsdaten-fuer-amerikanische-ki-die-computer-im-parlament-wurden-von-microsoft-eigenmaechtig-umgeruestet-ld.1917478
- Microsoft Learn, "Manage Microsoft 365 Copilot Chat," December 8, 2025. https://learn.microsoft.com/en-us/copilot/manage
- Microsoft Learn, "Enterprise data protection in Microsoft 365 Copilot and Microsoft 365 Copilot Chat," December 8, 2025. https://learn.microsoft.com/en-us/copilot/microsoft-365/enterprise-data-protection
- Microsoft Learn, "Shared responsibility in the cloud," December 3, 2025. https://learn.microsoft.com/en-us/azure/security/fundamentals/shared-responsibility
- Microsoft Learn, "Privacy and protections" (Microsoft Copilot Chat). https://learn.microsoft.com/en-us/copilot/privacy-and-protections
- Microsoft Learn, "Microsoft Copilot overview." https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-overview