Microsoft’s Telemetry Functionality in Windows 11 Raises Privacy Concerns
Windows 11 incorporates an embedded mechanism that tracks the utilization of various AI coding agents and developer tools. Should the optional diagnostic data feature be enabled, this information is relayed to Microsoft.
On November 16, 2026, security researcher Xusheng Li, affiliated with Binary Ninja, unveiled the initial discovery via a post on X. He subsequently authored a comprehensive technical analysis.
Insight into Windows 11 AI Agent Tracking
In a quest for undocumented functionalities, Li prompted an AI assistant to delve into recent files associated with Windows.
The investigation yielded a list of executable names within MissionControlAggregator.dll, including codex.exe, claude.exe, copilot.exe, and cursor.exe.
This compilation also encompassed widely used tools such as python.exe, dotnet.exe, gcc.exe, and code.exe. Upon inspecting the file through Binary Ninja, Li discerned that these identifiers were engaged in straightforward comparisons.
A designated feature flag termed CLICodingAgentUsage monitors these new evaluations.
Importantly, the DLL does not actively scan the user’s running applications. Rather, it registers events that Windows inherently generates, such as KernelProcess. AppStateChangeSummary and Win32kTraceLogging.AppInteractivitySummary.
Moreover, it is capable of observing Windows Subsystem for Linux (WSL) launch activities. Remarkably, even a standard Windows installation of codex.exe can trigger a match, indicating the checks extend beyond just Linux tools.
Upon identifying a match, the component sets a flag in its stored state. Subsequently, it transmits a report labeled Segments.SubCohorts via Event Tracing for Windows.
This report conveys binary flags—yes or no—yet omits details such as runtime, focus duration, or textual prompts. Despite its simplicity, a record of the AI tools employed by developers constitutes sensitive information.
Li scrutinized a Windows 11 ARM64 virtual machine operating on build 26200.9974. He discovered that this functionality was already activated, with the DiagTrack telemetry service operational.
DiagTrack initiated AggregatorHost.exe, which subsequently loaded the DLL. For testing purposes, he replicated cmd.exe and renamed it to codex.exe.
Intriguingly, the mere alteration of the name was sufficient to trigger the Codex flag. This action resulted in a report displaying CliAgentSubCohorts with a value of 4. Similarly, renaming gcc.exe activated an alternate developer tool flag.
Li extended his analysis beyond local events, recognizing that local logs alone are not definitive proof. He determined that a DiagTrack listener was attuned to the report. Furthermore, this report surfaced in DiagTrack’s upload queue as non-essential data.
This finding implies that Microsoft classifies it as optional diagnostic data. In a pristine installation of Windows 11 Home in the U.S., optional data was enabled by default, with AllowTelemetry configured to 3.
Conversely, a baseline virtual machine, sourced from the public ISO, had the coding-agent feature disabled. Consequently, it remains unverified whether this feature is universally active across all Windows PCs.
To modify this setting, navigate to Settings, select Privacy & security, and then access Diagnostics & feedback. Disable the option to Send optional diagnostic data.
According to Microsoft’s own privacy and diagnostics guide, your machine remains secure and fully functional with only the necessary data.
The guide also indicates that optional data may encompass device activity, which aligns with this scenario. For managed corporate devices, this setting might be concealed, prompting IT departments to review their specific policies.
AI coding tools are increasingly attracting the attention of malicious entities. Recent reports from Cyber Security News reveal that info-stealing malware is now targeting data pertaining to Claude, Cursor, and Codex stored on developer machines.
Additionally, the reports highlight emerging threats aimed at AI coding tools that exfiltrate credentials. Although a log of agent use is not a critical secret, it indicates where valuable code and tokens might be located.

Thus far, Microsoft has not clarified the rationale behind this data collection. It is plausible that such endeavors are intended for product development, as Windows continues to integrate additional agent functionalities.
In the absence of further clarification from the company, developers are advised to review their diagnostic configurations, consider Li’s findings, and determine their comfort level regarding data sharing.
Source link: Cybersecuritynews.com.




