Microsoft's relentless push to embed artificial intelligence across its entire product ecosystem is creating unintended consequences that undermine the company's core strength: a reliable operating system. Recent failures in Windows updates and controversial data-collection policies tied to Copilot development suggest that Satya Nadella's vision of AI-first computing may be moving faster than the engineering infrastructure can support .

What's Going Wrong With Windows Updates?

On March 26, 2026, Microsoft released KB5079391, a non-security update intended to improve Windows stability and reliability. The update was supposed to deliver "production-quality" improvements, including support for monitors beyond 1,000 Hz, a redesigned About page in Settings, and a brand-new Narrator built around Copilot . Instead, the update refused to install for many users, throwing up error code 0x80073712 in a loop, which indicates missing or corrupted files in the update package itself .

Microsoft pulled the update from its release channel after the installation failures became widespread. The company temporarily limited availability while investigating the issue, acknowledging that "rollout of this update is temporarily paused due to installation error 0x80073712" . Notably, no reports have emerged of the update actually breaking systems for users who managed to install it; the problem lies purely in the installation process itself. However, the optics damage is significant: this is yet another broken Windows update in a pattern that has eroded user confidence in the operating system's reliability.

The irony is sharp. Microsoft positioned this update as a stability improvement, yet it became a stability problem. The inclusion of Copilot-based features like the new Narrator suggests that AI integration was a priority in this release, raising questions about whether quality assurance processes kept pace with feature development .

How Is Microsoft Using Your Data to Train Copilot?

The Windows update failures are not isolated from a broader pattern of aggressive data collection tied to Copilot development. GitHub, which Microsoft owns, announced plans to begin using customer interaction data to train AI models starting April 24, 2026 . This policy change affects millions of developers worldwide and reveals how deeply Microsoft is mining user behavior to improve its AI systems.

GitHub's chief product officer Mario Rodriguez explained the rationale: "By participating you'll help our models better understand development workflows, deliver more accurate and secure code pattern suggestions, and improve their ability to help you catch potential bugs before they reach production" . The company noted that it has already incorporated Microsoft interactions to fine-tune model training processes, which have delivered marked improvements. Expanding the scheme to other users will support this at scale .

The data types GitHub plans to use for AI training include:

• Code Outputs: Code snippets that users accept or modify when using GitHub Copilot
• User Inputs: Inputs sent to GitHub Copilot, including code snippets and chat interactions
• Code Context: Code surrounding your cursor position and file names, repository structure, and navigation patterns
• Documentation: Comments and documentation written by users
• User Feedback: Interactions with Copilot features and thumbs up/down ratings on suggestions

GitHub emphasized that certain data types will not be used, including interaction data from Copilot Business, Copilot Enterprise, or enterprise-owned repositories . However, there is a critical caveat: while content from private repositories "at rest" won't be used, "Copilot does process code from private repositories when you are actively using Copilot," and this interaction data could be used for model training unless users opt out .

Who Is Affected and What Are Your Options?

The policy change applies to specific subscription tiers. Users on GitHub Copilot Free, Pro, and Pro+ accounts will be affected, while those on Copilot Business and Enterprise plans are exempt . Student and teacher accounts are also exempt from the new policy. However, affected users retain the ability to opt out.

Opting out is straightforward. Users can visit their settings at /settings/copilot/features, navigate to the "Privacy" section, and disable the option that reads "Allow GitHub to use my data for AI model training" . Rodriguez confirmed that users who previously opted out of product improvement data policies will retain those preferences automatically .

GitHub also clarified that data gathered as part of the program won't be shared with third-party AI model providers or independent service providers, though it "may be shared" with GitHub affiliates such as companies in Microsoft's broader corporate family . The company noted that it may "engage service providers to assist with model training" on its behalf, but only under the condition that this data is used "only for providing services to GitHub" .

Why This Matters for Nadella's AI Strategy

These two developments reveal the tension at the heart of Microsoft's AI ambitions under Satya Nadella's leadership. The company is racing to embed Copilot across its entire ecosystem, from Windows to GitHub to Office, treating AI integration as a competitive imperative. To make Copilot smarter and more useful, Microsoft needs vast amounts of training data, which it is now actively harvesting from millions of users across its platforms.

The broken Windows update suggests that this speed-to-market approach may be compromising the quality assurance that historically made Windows a reliable platform. When a stability-focused update fails to install due to missing files, it signals that the engineering organization may be stretched thin, prioritizing new AI features over the foundational reliability that enterprise and consumer users depend on .

Meanwhile, the GitHub data-collection policy reveals the scale of Microsoft's data-gathering operation. By training Copilot on real-world developer interactions, Microsoft is building a competitive moat around its AI capabilities. However, this approach also raises privacy and consent questions, particularly for users who may not fully understand that their code snippets, comments, and navigation patterns are being used to train models that Microsoft will monetize .

The challenge for Nadella is balancing innovation velocity with user trust. Windows' historical strength has been reliability; GitHub's strength has been trust among developers. If Microsoft's aggressive AI integration undermines either of those foundations, the company risks losing the very user bases that make its AI platforms valuable in the first place.