UK AI Weekly: Windows Is Watching Which AI Tools You Use
A Windows 11 component is recording whether developers use AI coding agents such as Codex, Claude and Copilot, then sending that information to Microsoft unless users opt out of optional diagnostic data.
That sounds less dramatic than βyour computer is secretly reading your code.β The evidence is narrower: Xusheng Liβs reverse engineering of Windows binaries found no prompt text or source code in the report. Instead, the system appears to track which tools were observed and encode those findings as bits.
Still, it matters.
Li traced the behaviour to MissionControlAggregator.dll, a component loaded by the DiagTrack Connected User Experiences and Telemetry service. On a Windows 11 ARM64 virtual machine running build 26200.9974, the relevant CLICodingAgentUsage experiment was already enabled, even though Li had not manually switched it on.
The component checks for four command-line agentsβcopilot.exe, claude.exe, codex.exe and agy.exeβand five desktop agents: github.exe, claude.exe, codex.exe, cursor.exe and antigravity.exe. It also recognises familiar developer tools including Python, .NET, GCC, Clang, Visual Studio and Visual Studio Code.
This is not limited to AI. The same machinery feeds broader categories covering programming languages, compilers and debuggers. Windows obtains application identities from existing operating-system events for native applications and WSL processes. A successful match sets a local state slot, which is later turned into a bitmap and emitted through Event Tracing for Windows as Segments.SubCohorts.
In other words, the report tells Microsoft that Codex was observed during a particular collection window. It does not necessarily tell Microsoft how long Codex was used, what prompts were entered or what code it handled. Those are meaningful boundaries, not a get-out-of-jail card.
For UK developers, the issue is less about whether Windows can recognise python.exe than why an operating system needs to catalogue the AI-assisted development stack in the first place. That question becomes sharper for organisations where employee tools interact with proprietary repositories, customer data or regulated systems.
The UK has no AI-specific statute and no single AI regulator. Existing bodies, including the ICO, apply the powers they already have. Here, the relevant framework is the UK GDPR and the Data (Use and Access) Act 2025. The source does not establish whether this telemetry breaches those rules, and it provides no response from Microsoft explaining the featureβs purpose, retention policy or consent design. So any claim that it is unlawful would be premature.
What it can trigger is familiar questions about transparency, lawful basis and whether telemetry is proportionate. The governmentβs pro-innovation stance, set out in the AI Opportunities Action Plan, does not make business-essential telemetry automatically acceptable. Innovation still has to survive basic trust.
What this means
This is not an AI safety story in the traditional sense. It is an operating-system trust storyβand those can be more immediately consequential. If Windows can distinguish whether a developer used an AI agent, it is also participating in a contest over which tools become the default route into serious work.
That creates pressure on Microsoft to explain the distinction clearly: security telemetry, optional diagnostics and product improvement should not be blurred together. For UK organisations, the sensible response is not panic, nor a fictional ban on Codex. It is to check what diagnostic settings exist, understand what enterprise deployments actually collect, and ask what evidence a proportional design would provide.
The one-sentence takeaway: Windows 11βs AI-tool telemetry may look modest on paper, but its reach makes clear why developer trust cannot be treated as an optional setting.
Source: Windows 11 is collecting data on which coding agents and developer tools you use β 636 points on Hacker News
Comments
Leave a message below. Your comment saves to your browser.