Copilot Trade-offs Surface

Microsoft is giving European customers more control over Microsoft 365 Copilot at the same moment it is creating a new escape hatch that can send some Copilot processing outside Europe when demand spikes. The change starts on April 17, 2026, and Microsoft says the setting is on by default for many eligible tenants. (learn.microsoft.com) The piece that can move is not the saved file in Word or Outlook. It is the “inferencing” step, which is the moment the large language model reads a prompt and generates an answer, like the few seconds when a waiter takes your order to the kitchen. (learn.microsoft.com) Microsoft says customer data at rest for Microsoft 365 Copilot still stays in the customer’s local region geography, and it says prompts and responses are included in those data residency commitments. The new flex routing policy changes where the live model run can happen during peak load, not where the stored interaction record normally lives. (learn.microsoft.com 1) (learn.microsoft.com 2) There is a catch in Microsoft’s own wording: when flex routing is allowed, associated pseudonymized data can also be stored outside the European Union data boundary for security and operational purposes. Microsoft says that data is encrypted in transit and at rest, but the promise is now “local storage, flexible processing” instead of a simple “everything stays here.” (learn.microsoft.com) This matters most for multinational companies that bought Copilot partly to satisfy internal rules on where employee and customer data can travel. A compliance team can accept encrypted overflow processing, but it still has to explain why an “European Union only” mental model no longer matches the operational design. (learn.microsoft.com 1) (learn.microsoft.com 2) At the same time, Microsoft is adding more knobs for administrators because Copilot is moving deeper into everyday work. Microsoft’s documentation says tenant administrators can change the flex routing setting in the Microsoft 365 admin center, and recent Copilot updates have focused on security, analytics, and automation controls for enterprise information technology teams. (learn.microsoft.com) (redmondmag.com) The infrastructure side of that trade-off showed up clearly at GitHub in March. In a report published on April 8, 2026, GitHub said it had four incidents in March that degraded service across GitHub products. (github.blog) The biggest incident hit on March 3, 2026, between 18:46 and 20:09 Coordinated Universal Time. GitHub said about 40% of github.com requests failed, about 43% of application programming interface requests failed, and GitHub Copilot requests saw an error rate of about 21%. (github.blog) GitHub traced that March 3 outage to a bug tied to its user settings cache, which caused every user’s cache to expire, recalculate, and rewrite at once. That surge created replication delays that spread into github.com, the application programming interface, Git operations, GitHub Actions, and GitHub Copilot. (github.blog) Two days later, on March 5, 2026, GitHub Actions was degraded for 2 hours and 55 minutes after Redis load balancer configuration changes sent internal traffic to the wrong host. GitHub said 95% of workflow runs failed to start within 5 minutes, the average delay reached 30 minutes, and 10% of workflow runs failed with an infrastructure error. (github.blog) GitHub’s report says “deep architectural work” is already underway, along with faster fixes like a killswitch, better monitoring, and moving the cache mechanism to a dedicated host. That is the other side of the Copilot pitch: the assistant feels personal inside Word, Outlook, or Visual Studio Code, but the system underneath is still a shared cloud service that can wobble for everyone at once. (github.blog)