A researcher found Windows 11 checks for claude.exe, codex.exe, cursor.exe and copilot.exe and reports which ones you run. Here's how to check and disable it.
Vamsi Tallapudi
Manager, Architect Technology at Cognizant
Windows 11 has a built-in component that checks whether you run Claude Code, Codex, Cursor, GitHub Copilot, or other coding agents, then quietly reports which ones it found. It's not a rumor: a researcher reverse-engineered the exact DLL doing it, and the behavior is confirmed on a current Windows 11 build. Here's how to check if it's on and how to turn it off.
The feature was found by Xusheng Li, who publishes detailed Windows internals research at xusheng.dev. In a post dated November 16, 2026, Li traced a component called MissionControlAggregator.dll, loaded by AggregatorHost.exe, which is itself spawned by DiagTrack, the Connected User Experiences and Telemetry service that's been part of Windows since Windows 10.
The DLL subscribes to two existing ETW (Event Tracing for Windows) providers: KernelProcess.AppStateChangeSummary and Win32kTraceLogging.AppInteractivitySummary. These aren't new. Windows has used them for years to track which apps you run and for how long. What's new is what it's matching against. On the build Li tested (26200.9974, with MissionControlAggregator.dll version 10.0.26100.9818), the component checks process names against a specific list of coding tools:
codex.exe, claude.exe, copilot.exe, agy.exegithub.exe, claude.exe, codex.exe, cursor.exe, antigravity.exepython.exe, gcc.exe, dotnet.exe, code.exe, and othersWhen a match hits, Windows flips a bit in a local state file. Roughly once a day, it packages those bits into a single ETW event named Segments.SubCohorts and queues it for upload. The whole thing sits behind a feature flag with a name that doesn't leave much to the imagination: CLICodingAgentUsage, feature ID 62580080.
The researcher behind the account Crackmes.one independently confirmed the same mechanism and listed matching the same executable set:
No. Li checked for this specifically and found none. The report is a bitmap: it tells Microsoft "claude.exe ran on this machine," not what you typed into it or what Claude Code did with your repo. Detection is also dumb in a useful way: it matches on executable filename only. Li tested this by renaming cmd.exe to codex.exe and the component flagged it as "Codex" anyway. So it's not inspecting binaries or verifying signatures. It's reading the process name off the kernel event and checking it against a list. Cheap, fast, and also trivially spoofable in either direction.
Both, depending on what you mean. The underlying telemetry pipe (DiagTrack, ETW process tracking, app usage events) has existed since Windows 10 and is well documented in Microsoft's own diagnostic data guide. Microsoft has always collected "app activity, such as which programs are launched on a device, how long they run" under its Optional diagnostic data tier. That's not a secret; it's in the official data-collection table.
What's new is the specific target list. Someone on the Windows telemetry team added coding agents and AI dev tools as named entries worth tracking, with their own feature flag, sometime in the last few builds. That's a deliberate decision, not a side effect of generic app tracking. It tells you Microsoft cares, at a product-planning level, about which AI coding tools are gaining share on Windows machines, which tracks with Microsoft pushing its own Copilot and the Windows AI agent features hard this year.
Per Microsoft's documentation, there are four diagnostic data levels:
| Level | Value | What's included |
|---|---|---|
| Diagnostic data off (Security) | 0 | Nothing sent (Enterprise/Server/Education only) |
| Required (Basic) | 1 | Minimum device health and compatibility data |
| Enhanced | 2 | Windows 10 only, deprecated on Windows 11 |
| Optional (Full) | 3 | Adds app activity, browsing data, enhanced crash reports |
App-launch tracking (the category the coding-agent check falls under) only ships when a device is set to Optional diagnostic data. Li found this was the default on a US Windows 11 Home setup. If your machine is set to Required diagnostic data, this specific report shouldn't leave your PC at all, because the "app activity" category simply isn't included at that tier.
You don't need special tools to check. Open PowerShell and run:
Get-Service DiagTrack | Format-Table Name,Status,StartType -AutoSize
Get-Process AggregatorHost -ErrorAction SilentlyContinue
If DiagTrack shows Running, the telemetry pipe is active. AggregatorHost.exe showing up in the process list means MissionControlAggregator is loaded. Neither of these alone proves the coding-agent flag is on, since that depends on your diagnostic data tier and the CLICodingAgentUsage flag's rollout state. But it confirms the machinery exists on your machine.
The simpler, no-PowerShell check: go to Settings → Privacy & security → Diagnostics & feedback and look at what's set under diagnostic data. If "Send optional diagnostic data" is on, assume app-launch tracking, including this coding-agent list, is active.
That one toggle drops you to Required diagnostic data, which (per Microsoft's own documentation) excludes app-activity reporting entirely. It's also the single lever available on Windows 11 Home. If you're on Pro, Enterprise, or Education, you get more granular control:
Computer Configuration → Administrative Templates → Windows Components → Data Collection and Preview Builds → Allow diagnostic data, set to 1 (Required) instead of 3 (Optional)System/AllowTelemetry policy CSP, same valuesSystem/LimitDiagnosticLogCollection to stop diagnostic logs specifically, separate from the event-level reportingNone of this is a registry hack or an unsupported workaround. It's the documented, Microsoft-sanctioned diagnostic data setting. The only thing unusual here is what it was quietly expanded to cover.
I get the instinct to shrug this off: it's a bitmap, not your code. But think about what the bitmap actually enables. Microsoft now has a documented, flagged mechanism to answer "how many machines in region X run Claude Code vs. Codex vs. Copilot vs. Cursor" at scale, continuously, without asking anyone. That's market intelligence on AI coding tool adoption, collected through the OS vendor's own telemetry pipe, while two of the four tools on that list (Copilot and GitHub, aka Copilot CLI) are made by the same company running the pipe. Nobody needs a conspiracy theory for that to feel uncomfortable. It's a structural conflict of interest, flag or no flag.
And the Hacker News thread discussing Li's post is a decent gut check on how people reacted, worth skimming even though the comments wander (ours did, into a Linux-gaming tangent, the way HN threads do).
If you've been keeping tabs on usage limits and cost tracking for Claude Code, this is a different angle on the same theme: see our breakdown on Claude Code usage limits and cost for the token-side accounting, and ChatGPT and Codex's shared usage limit for how OpenAI handles the same problem on its side. If you're choosing between agents in the first place, our Claude Code vs. Codex vs. Cursor comparison is still the most relevant read before you pick one.
This isn't Windows reading your prompts or your source files. It's Windows counting which AI coding tool executables ran on your machine and phoning that home when you've opted into optional diagnostics, which, on a default US Home install, you probably have, whether you remember agreeing to it or not. One toggle in Settings turns it off. If you manage machines for a team, the Group Policy and Intune equivalents give you the same control at scale. Either way, it's worth five minutes to go check, especially if "which coding agent I use for work" isn't information you want sitting in a telemetry pipeline owned by a company that also sells a competing coding agent.
New AI tools, automation workflows, and course drops — straight to your inbox. Join 2,400+ builders.
CMU's free CS Learn platform puts 4 full courses on YouTube, including an AI agents class. Here's what each covers and the order to actually watch them in.
Qwen4-Max left preview on November 16 at $1.50/$4.50 per million tokens. Here's how its real benchmarks and coding tests stack up against Claude Opus 5.5.
Google Antigravity's /teamwork needs a paid plan and can flood your folder with files. Claude Code subagents are cheaper but need manual orchestration. Here's which wins for long projects.
Comments