Skip to content

Signals · Power

Windows 11 is recording which AI coding agents you run, and in one US test the upload setting was on

A reverse-engineered Windows 11 component flags Claude, Codex, Copilot and Cursor use and queues the record for Microsoft when optional diagnostics are on.

by ·3 min read·

New here? Start with the free AI Survival Kit →

Windows 11 is recording which AI coding agents you run, and in one US test the upload setting was on

A reverse-engineering write-up published this week by Xusheng Li on xusheng.dev shows that Windows 11 contains a component that records which AI coding agents and developer tools run on a machine. When optional diagnostic data is on, Windows queues that record for upload to Microsoft. The component is MissionControlAggregator.dll. It checks for copilot.exe, claude.exe, codex.exe and agy.exe as command-line agents, and for github.exe, claude.exe, codex.exe, cursor.exe and antigravity.exe as desktop agents. The feature that controls those checks is named CLICodingAgentUsage, feature ID 62580080. The same component also looks for node.exe, python.exe, dotnet.exe, gcc.exe, clang.exe, javac.exe and code.exe, among others.

What the evidence shows

The report records which tools ran. Li found no prompt text or source code in it. The component emits a report called Segments.SubCohorts, a set of bits covering twelve broader tool categories, four command-line agents and five desktop agents. The Windows events it reads from include runtime, time in focus and interaction counts. According to the write-up, the component appears to discard those details and keep only which agents were seen in the reporting window.

For the Codex check, the file name alone was enough. Li copied cmd.exe to a file called codex.exe and ran it. Windows set the Codex bit. A copy named gcc.exe set the compiler and debugger category.

In one US test setup, the upload setting was on by default. On a clean install of Windows 11 Home ARM64 from the latest official ISO, with the region set to the United States, the Diagnostic data toggle was already enabled and labelled Include Optional. After setup, AllowTelemetry was 3, which confirmed it.

The write-up also gives a limit, and it matters. The default was checked in one US setup. The agent-usage event was found on Li's Parallels Windows 11 ARM64 VM, where CLICodingAgentUsage was already enabled as a service experiment and Li had not enabled it manually. There Li found the event in the upload queue of DiagTrack, the Windows telemetry service, waiting to be sent to Microsoft.

Why a bitmap still matters

One machine's bitmap is small. Each report says which coding agents a machine ran in that window. The write-up does not say what Microsoft does with the reports once they arrive. It sits inside the operating system's own telemetry.

For an operator, the point is narrower and more useful. Your choice of tools is now data about you, collected below the level where you would notice it. The layer that keeps its value as AI spreads includes judgement about your own setup: what each tool can see, what it sends, and who receives it. AI does the routine work. You do the thinking, and that includes deciding what your machine says about how you work.

Li's position is that collecting which agents and developer tools someone uses already has significant privacy implications, even though Li found no prompt text or source code in the report. That is a decision about defaults. In Li's US test, the install screen arrived with it already made. You can make it again.

Your Next Move

  1. Check the setting today. Open Settings → Privacy & security → Diagnostics & feedback and decide on Send optional diagnostic data on purpose. Li confirmed the default in a US setup by finding AllowTelemetry set to 3. If you never changed it during setup, assume it is on.
  2. Add telemetry to your team's agent policy. If you approve which coding agents your team may use, put the operating system's diagnostic level in the same document. Choosing an agent and choosing what reports on that agent are one decision, so one owner should make both.
  3. Map your stack on one page. For each agent and developer tool you run, write down what it can see, what it sends, and who receives it. Where you cannot fill a row in, that row is your next piece of research.

I build systems like the one publishing this site. → Work with me

About the author

Jo

Jo runs The War Room: one signal a day on how AI is changing work, and what to do about it.

Sources

Get the Briefing

Want more intelligence like this?

The weekly briefing, free — and the AI Survival Kit with it.

Takes 30 seconds. No spam.

More in Signals

See all