Microsoft's Three Decades of Data Capture: From Halloween Documents to Copilot

Forum topic · 小凯 · 2026-05-06

Summary

This Chinese forum post traces Microsoft's evolving relationship with open source over thirty years, arguing the company has moved from demonizing open source (the 1998 'Halloween Documents' era), to embracing it, to now absorbing developer data at scale. The author contends that GitHub's 2026 terms update, which distinguishes between 'code at rest' (owned by the developer) and 'interaction data' (becoming raw material for training Microsoft's AI models), constitutes 'Embrace, Extend, Extinguish 2.0'. The post cites the Recall controversy of 2024, where screen recordings were reportedly stored in plaintext SQLite databases, as evidence of full-spectrum data capture. It argues developers have traded digital sovereignty for modest productivity gains, and closes with a reference to Linus Torvalds' early-2000s remarks about Microsoft, suggesting today's threat is friendlier but more pervasive: Copilot sits in the passenger seat, quietly recording every keystroke.

Thirty years on, Microsoft's attitude toward open source has traced a perfect Möbius strip: from cursing it as a cancer, to professing love for it, to attempting to swallow it whole. If the leaked "Halloween Documents" of 1998 exposed a walled garden built on technical barriers, then the 2026 GitHub Copilot terms update marks the garden's evolution into a full-time, always-on data-capture engine.

From Code Ownership to "Interaction Data"

The core of this evolution is Microsoft's seizure of the power to define "developer productivity." Back in 2004, when Linus Torvalds mocked Microsoft funders, the flashpoint was still code ownership. By 2026, the very concept of ownership has blurred. According to GitHub's latest updated terms, code is subdivided into:

Code at rest: when you merely store code, it remains yours.
Interaction data: the moment you open it in VS Code, edit it, or trigger AI completion, it legally becomes raw material for training Microsoft's AI models.

This definitional shift—from "storage" to "interaction"—is Embrace, Extend, Extinguish (EEE) 2.0.

Recall and Full-Spectrum Capture

Microsoft no longer tries to stop you from running Linux. It even encourages development inside WSL on Windows—because as long as you stay on the Windows desktop, it can capture everything through the Recall feature. Recall's 2024 security scandal—storing user screen captures in plaintext in a SQLite database—revealed the mechanism's essence: not an efficiency tool, but a digital tape recorder that bypasses every privacy line of defense.

The bargain developers made

Microsoft's open stratagem: it has convinced developers worldwide that a 5% productivity gain is worth surrendering 100% of their digital sovereignty. Through the GitHub acquisition and the free distribution of VS Code, Microsoft has established an irreversible chain of dependency. Developers find themselves in a "digital comfort zone": the best autocomplete, the smoothest deployment pipelines—but the price is that every line of logic, every innovation, is being ground up in real time and recombined into Microsoft's private assets.

Torvalds' coldest prophecy

When Torvalds insulted his rival's integrity twenty years ago, he was protecting the purity of code's lineage. Today's opponent has superior technical taste and a far gentler demeanor. It no longer points at you and curses—it sits in your "copilot" seat, smiling, recording every skid mark. Three decades of this hunt teach a brutal fact: in the digital world, there are no permanently free hotel rooms, only tenancy at a steep price.