Windows 11 Checks Whether You Ran Claude, Codex or Copilot, and Queues a Report for Microsoft
Software / news
Windows 11 Checks Whether You Ran Claude, Codex or Copilot, and Queues a Report for Microsoft
A researcher's build 26200.9974 teardown found a bitmap of coding tools seen on the machine, uploaded under optional diagnostics, with no Microsoft statement in the coverage.

A Windows 11 diagnostics component checks whether Claude, Codex, Copilot, Cursor and other coding tools have run on a machine, and queues a report to Microsoft when optional diagnostic data is on, according to a Nov. 16 reverse-engineering write-up by researcher Xusheng Li. Neither source consulted includes a Microsoft statement.
Li's post found that the report is a bitmap of which tools were seen, not a log of what they did. That limits the privacy cost, but it does not make the collection invisible. Most of what follows rests on one researcher's measurements, which is why each claim below carries his name.

What Li found inside the Windows build
Li tested a Parallels Windows 11 ARM64 virtual machine on build 26200.9974. The component is MissionControlAggregator.dll, version 10.0.26100.9818, which runs inside AggregatorHost.exe, a child of the DiagTrack service. The feature is named CLICodingAgentUsage, feature ID 62580080, and he reports it was already enabled as a service experiment.
It does not scan running processes. Li says it subscribes to events Windows already emits: KernelProcess.AppStateChangeSummary and Win32kTraceLogging.AppInteractivitySummary for Windows programs, and Lxss exec events for WSL. He renamed cmd.exe to codex.exe and wrote that "The name alone was enough for the component to recognize it as Codex."
Which tools it looks for
| Category | Executables named in the post |
|---|---|
| CLI agents | copilot.exe, claude.exe, codex.exe, agy.exe |
| Desktop agents | github.exe, claude.exe, codex.exe, cursor.exe, antigravity.exe |
| Developer tools | node.exe, python.exe, dotnet.exe, gcc.exe, clang.exe, gdb.exe, javac.exe, devenv.exe, code.exe |
The post lists these "among others," and says some checks also verify the publisher.
What gets recorded and sent
Each tool category has a bit. A match sets the bit and emits a Segments.SubCohorts event. In Li's test, running the renamed Codex binary produced CliAgentSubCohorts: [4], which is 1 << 2, and a gcc.exe test produced SubCohorts: [8], which is 1 << 3.
Li writes that the component "appears to discard" runtime, focus time and interaction details, keeping "only a bitmap of which agents were observed," and that he found no prompt text or source code in the report. Local state is stored under C:\ProgramData\Microsoft\Diagnosis\AggregatorStorage\MissionControlAggregator$.
On upload, Li says "Windows queues the coding-agent usage report for upload to Microsoft" when optional diagnostic data is allowed, and the event is marked "isRequired": false. He confirmed it in DiagTrack's upload queue.
It was on by default in his test
On a clean install of Windows 11 Home ARM64 from the latest ISO, with the US region, the diagnostic data toggle already sat at "Include Optional" and AllowTelemetry was 3. To turn the report off, decline optional diagnostics during setup, or switch off "Send optional diagnostic data" under Settings, Privacy & security, Diagnostics & feedback.
Borncity republished the finding on Nov. 17 and reports the same path. It adds that administrators can set the policy through Group Policy or mobile device management. Borncity states its article was created with AI assistance and reviewed editorially. It also notes that Li measured only a US default install, and no measurements exist yet for German installations.
What is not established
Three things are open. Microsoft has not been quoted explaining what the data is for. The test covered a single build in a virtual machine, so whether the feature reaches every Windows 11 install is unknown. And Li describes the feature as a service experiment, which would let Microsoft change its behaviour without shipping an update, though the post does not say so directly.
The step Li gives is to set optional diagnostics off. Sandboxes such as Maritime's agent microVM isolate the agent itself, but the post describes detection at the host operating system, which a guest VM does not control. Build machines are the same case, including those that run Google's ABFS checkout tooling.
Sources
More in Software
- 01OpenAI's WebRTC Voice Stack Draws a 30 ms Objection That Its Own Engineers Never AddressedLuke Curley says WebRTC drops audio older than 30 ms. OpenAI's engineers wrote about port ranges and routing, which is a different problem.
- 02Superpowers Passes 322,000 GitHub Stars as a Set of 15 Prompted Skills for 16 Coding AgentsJesse Vincent's MIT-licensed framework tells agents to brainstorm, plan and test first. Its clone count is large, and its evidence for better results is one author's view.
- 03GitHub Disables 144 Picofly Repositories After a Nintendo Notice Names OneThe November 6 notice is the first on GitHub aimed at Switch modchip firmware, and it asked for forks to be taken down without listing a single one.
- 04Chrome's Gemini Nano Download: What Google's Docs Say, and the Numbers They Leave OutGoogle's page on built-in model management gives triggers, a 30-day rule and full-download updates, but no file size, no disk threshold and no consumer off switch.