Windows 11 Coding Agent Telemetry: Opt Out [2026]
A reverse engineer proved Windows 11 logs which AI coding agents you run — claude.exe, cursor.exe, copilot.exe — and uploads it to Microsoft by default. Here's how to turn it off.
Windows 11 coding agent telemetry is a newly-discovered system feature that records which AI coding tools you run — claude.exe, cursor.exe, copilot.exe, codex.exe and more — writes that usage to a local diagnostics report, then uploads it to Microsoft through the "optional" diagnostic data channel that ships enabled by default.
The finding is less than a day old as I write this. It hit 410 points and 260-plus comments on Hacker News within about seven hours, and until now nobody had published how to actually turn it off.
That's the gap this post fills. Independent researcher Xusheng Li did the hard part — pulling MissionControlAggregator.dll apart in Binary Ninja and tracing the executable-name checks to a live Windows component. His writeup is excellent reverse engineering. It is not a how-to. If you use any AI coding agent on Windows 11, you want the how-to.
Key takeaways
- Windows 11 ships a system component,
MissionControlAggregator.dll, that string-matches coding-agent executables likeclaude.exe,cursor.exeandcopilot.exeand logs their usage.

- The data is written locally first, then uploaded to Microsoft via optional diagnostic data — which is on by default on a fresh Windows 11 install.
- You can stop the upload in under two minutes: Settings > Privacy & security > Diagnostics & feedback, and turn off "Send optional diagnostic data."
- Pro, Enterprise and Education admins get a stronger lever through Group Policy, the
AllowTelemetryregistry key, and Intune. - This is a third, undisclosed telemetry layer that sits on top of whatever Claude Code, Cursor and GitHub Copilot already collect themselves.
What Microsoft's MissionControlAggregator.dll actually tracks
Windows 11 coding agent telemetry lives inside a diagnostics component called MissionControlAggregator.dll. The DLL contains hardcoded string comparisons against the names of specific coding-agent executables, groups them into usage "slots," records which ones have run on the machine, and surfaces that into a Windows Diagnostics report built on Event Tracing for Windows (ETW).

The whole thing is gated behind an internal feature flag named CLICodingAgentUsage, feature ID 62580080. That naming matters: a numeric feature ID in that range is a staged, flighted feature, not something that's been quietly baked into Windows for a decade. It's new, and it's being rolled out through Microsoft's normal feature-flag machinery, which means it can expand to more machines without a visible update.
Telemetry you can't see isn't diagnostics. It's surveillance with a nicer label. Log locally if you must, but never upload by default.
The reason this landed so hard is context. Microsoft's Windows Recall feature — which periodically screenshots everything you do — drew sustained privacy backlash after security researcher Kevin Beaumont dissected how it stored sensitive data in plaintext. A second, quieter AI-usage hook arriving in the same OS, aimed squarely at developers, was never going to be received warmly. I've written before about how Claude Code handles its own telemetry, and the pattern is identical: the collection is defensible on paper and alarming in practice because nobody consented to it explicitly.
Which coding agents and dev tools are on the list
Xusheng Li's disassembly shows two sets of checks. There are four CLI-agent slots matching copilot.exe, claude.exe, codex.exe and agy.exe, with extensionless equivalents for the WSL route — so a coding agent invoked inside Windows Subsystem for Linux is covered too. Then there are five desktop-agent slots checking github.exe, claude.exe, codex.exe, cursor.exe and antigravity.exe.

The list doesn't stop at AI agents. The same component checks for general developer tools: node.exe, python.exe, dotnet.exe, gcc.exe, clang.exe, gdb.exe, javac.exe, devenv.exe and code.exe. And it's not naive about it. Some checks verify publisher metadata to resist spoofing — the Node.js slot expects a Node.js Foundation signature, Python expects Python Software Foundation, and the .NET and Visual Studio slots expect Microsoft Corporation. Rename your shell to python.exe and it won't fool the matcher.
That code.exe check is the one with the widest blast radius. VS Code is used by 75.9% of developers per the 2025 Stack Overflow Developer Survey of more than 49,000 respondents — more than double its nearest rival. So this OS-level fingerprint touches the large majority of working developers whether or not they've ever opened an AI agents workflow. Meanwhile Anthropic's terminal agent sits at 155,944 GitHub stars, which tells you how many machines a claude.exe match could plausibly fire on.
There's a special awkwardness here worth naming: copilot.exe is on the list, and GitHub Copilot is a Microsoft product. Microsoft is simultaneously the OS vendor doing the tracking and the vendor of one of the tracked tools. If you want the head-to-head on those agents, I've compared GitHub Copilot vs Claude Code and Cursor vs Claude Code elsewhere.
How the data gets to Microsoft
The pipeline is two stages, and the distinction is the whole ballgame. First, MissionControlAggregator.dll detects a matching executable and writes the finding into a local Windows Diagnostics ETW report. Second, that report gets uploaded to Microsoft — but only if optional diagnostic data is enabled.
On a fresh Windows 11 install, optional diagnostic data is enabled by default. So the upload happens unless you've gone and turned it off, which almost nobody does during setup. The transport is the Connected User Experiences and Telemetry service, known by its service name DiagTrack. That's the background service that batches diagnostic events and ships them to Microsoft's endpoints.
What Windows calls "required" diagnostic data is the minimum the OS insists on sending. "Optional" diagnostic data is the superset that includes app usage, browsing activity in some cases, and — now — coding-agent usage. The coding-agent report rides in that optional bucket, which is the lever you'll pull in a minute.
How to verify this on your own machine
You don't need Binary Ninja or reverse-engineering chops to confirm diagnostic data is flowing. Three tools get you there:
- Diagnostic Data Viewer. Turn it on under Settings > Privacy & security > Diagnostics & feedback > "View diagnostic data," install the app from the Microsoft Store, and browse the actual events your machine has queued or sent. You can search and filter the raw telemetry Windows holds about you.
- Event Viewer. The ETW reports that feed diagnostics are visible in the Windows Event Log under the diagnostics and Operational channels if you want to watch the plumbing.
- Process Monitor. Mark Russinovich's Sysinternals Process Monitor lets you watch the
DiagTrackservice and the aggregator component touch files and registry keys in real time, which is how you'd confirm it's live rather than dormant.
If you'd rather trust a reproducible method than a screenshot, that's the toolkit. I lean on the same observe-the-process discipline in the site's reverse-engineering writeups — watch what the binary does, not what the docs claim.
How to opt out of Windows 11 coding agent telemetry
Here's the step-by-step. Pick the row that matches your edition; the first option covers Home and Pro, the rest are for managed or power-user setups.
- Settings UI (Home and Pro). Open Settings > Privacy & security > Diagnostics & feedback. Under "Diagnostic data," turn off Send optional diagnostic data. This is the single switch that stops the coding-agent report from leaving your machine. Two minutes, no reboot.
- Group Policy (Pro, Enterprise, Education). Run
gpedit.msc, then navigate to Computer Configuration > Administrative Templates > Windows Components > Data Collection and Preview Builds > Allow Diagnostic Data. Enterprise and Education can set this to Diagnostic data off. Pro can go no lower than Send required diagnostic data, which still drops the optional coding-agent payload. - Registry (all editions). Set
HKLM\SOFTWARE\Policies\Microsoft\Windows\DataCollection\AllowTelemetryto0as a DWORD. A value of0is "Security/Off" and is honored fully on Enterprise and Education;1is "Required." This is the same knob Group Policy writes under the hood. - Disable the DiagTrack service. Open
services.msc, find Connected User Experiences and Telemetry, stop it, and set Startup type to Disabled — or runsc config DiagTrack start=disabledfrom an elevated prompt. This kills the upload transport outright. It's the heaviest hammer and can affect a few diagnostics-dependent features, so use it deliberately. - Intune (managed fleets). Push the diagnostic data configuration through the Policy CSP and set the allowed level to Diagnostic data off / level
0across the fleet, so individual devs can't accidentally re-enable it.
For a visual walkthrough of stripping the broader Copilot and AI surface out of Windows, this covers the territory:
Does turning off diagnostic data stop local logging, or just the upload?
This is the question everyone asks second, and the honest answer is: turning off optional diagnostic data stops the upload, not necessarily the local detection. The two-stage design means the aggregator component can still run its executable-name checks and write findings to the local ETW report; what the setting controls is whether that report leaves your machine.
In practice, with optional diagnostic data off, Microsoft isn't receiving the coding-agent fingerprint. For most people that's the outcome that matters. If you want to be certain nothing local is even being recorded, you move to disabling the DiagTrack service entirely (step 4 above), since that's the collector. I'd treat "stop the upload" as the pragmatic default and "disable DiagTrack" as the privacy-maximalist option.
This mirrors how I think about telemetry in my own systems. I built and run the seven-agent pipeline that researches, writes and publishes this blog — 261-plus posts to date — and one rule I enforce in it is simple: log plenty locally for debugging, but gate every outbound call behind an explicit opt-in key. Windows inverts that default. That inversion, not the logging itself, is the whole problem. Log locally if it helps you; never ship by default.
Tool-by-tool: Claude Code, Cursor, and GitHub Copilot already collect too
Here's the part the original writeup skips. The Windows layer is a third telemetry stream. Each agent already phones home on its own. The new finding doesn't replace that — it adds a cross-tool usage fingerprint at the OS level that Microsoft specifically didn't have before.
| Layer | What it collects | On by default? | How to control it |
|---|---|---|---|
Windows 11 OS (MissionControlAggregator.dll) | Which coding agents / dev tools ran, via executable-name match | Yes (optional diagnostic data) | Settings toggle, Group Policy, registry, DiagTrack |
| Claude Code (Anthropic) | Code accept/reject signals, conversation data, /bug feedback; stated policy against training on feedback | Yes, with limited retention | Env vars / privacy settings in the CLI config |
| Cursor (Anysphere) | Usage and editor telemetry; Privacy Mode option that stops code retention | Partial | Enable Privacy Mode in settings |
| GitHub Copilot (GitHub/OpenAI/Microsoft) | Prompt/suggestion telemetry; enterprise admin content-exclusion controls | Yes, with admin overrides | Org admin policy / content exclusions |
The nuance that changes the risk calculus: your Claude Code usage data goes to Anthropic, your Cursor data goes to Anysphere, your Copilot data goes to GitHub. Before this finding, no single vendor saw which agents you switch between. The OS-level check hands Microsoft exactly that — a map of your whole AI toolchain, assembled on the one platform you can't easily avoid. If you're weighing Copilot vs Cursor, note that the OS layer logs both regardless of which you pick.
What IT admins and regulated teams should do
If you run a fleet of managed Windows 11 machines and your developers use AI coding agents, this is a compliance question, not a hobby question. A SOC 2, GDPR, or HIPAA shop now has an OS component recording tool-usage metadata and uploading it to a third party by default. That's the kind of undisclosed data flow auditors ask about.
Three moves for admins:
- Set the fleet-wide diagnostic data level to off (Enterprise/Education) via Intune or Group Policy, so the optional bucket — and the coding-agent payload with it — never ships.
- Document it in your data-flow inventory. If you process regulated data and your devs run agents against it, the OS telemetry belongs in your record of processing activities until Microsoft clarifies what's retained.
- Decide your agent policy explicitly. The same AI security discipline you apply to prompt injection and permission scoping applies here — know what each tool sends and where. I walk through the regulatory side of this in the guide to EU AI Act compliance for AI agents.
The boring answer is the right one here: treat OS telemetry as an untrusted data egress and gate it at the fleet level before a single developer logs in.
Is this Windows 11-only, and has Microsoft responded?
The finding is confirmed in Windows 11 binaries. The CLICodingAgentUsage flag is a flighted feature, so its exact reach across stable versus Insider channels isn't fully mapped yet, and there's no public confirmation it exists in Windows 10 or Windows Server. Treat "Windows 11, current builds" as the confirmed scope and watch for build-specific reports.
As of publication, Microsoft has not issued a public response and the behavior isn't documented in the Microsoft Privacy Statement or diagnostic data reference. That's the real story: a developer-targeted usage check, shipped under a flag, uploaded by default, and absent from the docs. If Microsoft updates the diagnostic data documentation to disclose it, that'll be the tell that it's becoming a permanent fixture rather than an experiment.
The cleanest escape hatch is the one the Hacker News thread kept circling back to: run your AI coding workflow somewhere the OS vendor isn't also your tool vendor. A Linux box, or a local LLM setup where the model and the agent both live on hardware you control, removes this entire question. Based on the local-LLM benchmark data I maintain at kunalganglani.com/llm-benchmarks, capable coding models now run on a single consumer GPU or an Apple Silicon machine, so moving your agent off Windows no longer means downgrading the model. That's not a slogan — it's the architecture choice that makes the telemetry moot. The developers who prize privacy enough to disable five Windows settings are the same ones who'll quietly move their vibe coding off Windows entirely, and Microsoft just handed them a fresh reason.
Photo by Joshua Hoehne on Unsplash.
Kunal Ganglani (2026, November 17). Windows 11 Coding Agent Telemetry: Opt Out [2026]. Kunal Ganglani. Retrieved November 17, 2026, from https://www.kunalganglani.com/blog/windows-11-coding-agent-telemetry
Frequently Asked Questions
How do I stop Windows 11 from sending diagnostic data to Microsoft?
Open Settings > Privacy & security > Diagnostics & feedback and turn off "Send optional diagnostic data." That stops the optional bucket, which is where the coding-agent usage report rides. For a harder block, Pro and Enterprise can use Group Policy or set the AllowTelemetry registry key to 0, and you can disable the Connected User Experiences and Telemetry (DiagTrack) service entirely.
Does turning off diagnostic data stop local logging or just the upload?
It stops the upload. The Windows component can still run its executable-name checks and write findings to a local ETW report; disabling optional diagnostic data just prevents that report from being sent to Microsoft. If you want to stop local collection too, disable the DiagTrack service, which is the actual collector and transport.
Is Cursor or Claude Code sending my code to Microsoft through this?
No. The Windows component only records that an executable like cursor.exe or claude.exe ran — a usage fingerprint, not your source code. Your actual code and prompts still go to Anthropic, Anysphere, or GitHub through each tool's own separate telemetry. The new Windows layer adds a cross-tool usage map Microsoft didn't have before, not the contents of your work.
What is the CLICodingAgentUsage feature flag?
It's the internal Windows feature flag (ID 62580080) that gates the coding-agent detection in MissionControlAggregator.dll, found by reverse engineer Xusheng Li. A numeric flag in that range indicates a staged, flighted feature rather than a long-standing one, meaning Microsoft can expand its rollout without a visible update. It's confirmed in current Windows 11 builds.
Can I fully disable Windows 11 telemetry without Group Policy?
On Home edition you can turn off optional diagnostic data in Settings and disable the DiagTrack service through services.msc, which covers the practical cases. You cannot reach the Enterprise-only "Diagnostic data off" level without Group Policy or the registry AllowTelemetry=0 key, but disabling DiagTrack stops the upload transport regardless of edition.
What should IT admins do about this on managed fleets?
Set the diagnostic data level to off fleet-wide via Intune or Group Policy so the optional coding-agent payload never ships, document the data flow in your compliance inventory if you handle regulated data, and define an explicit agent-usage policy. Treat the OS telemetry as an untrusted egress and gate it before developers log in.



