NTLite version 2026.04.10936 has arrived, and it hands users a sharp new knife: the ability to carve AI-related components out of Windows 11 25H2 installation images before they ever touch a hard drive. The update lands as frustration over Microsoft’s aggressive AI integration continues to simmer across forums, Reddit, and enterprise IT departments. For the first time, the popular Windows customization tool explicitly lists components like Copilot, Recall, and Windows Studio Effects as removable during image editing, giving privacy-conscious users and system administrators a cleaner baseline.
The timing is no accident. Windows 11 25H2, the feature update expected to ship in the second half of 2025, deepens the operating system’s reliance on on-device AI models. Copilot is woven deeper into the shell, Recall has evolved from its rocky preview into an always-on photographic memory for your desktop, and background AI services handle everything from live captions to adaptive power management. For many, this represents convenience. For others, it’s bloatware that hoovers up CPU cycles and raises uncomfortable questions about data collection. NTLite’s latest build cuts that tension by letting anyone craft an AI-free Windows install.
What NTLite Actually Does
NTLite is a configuration tool for Windows images. It mounts an ISO or an existing installation, presents a tree of editable components, and lets you remove features, slipstream updates, integrate drivers, and tweak settings before deployment. The tool has been a staple of power users and IT pros for over a decade, evolving alongside each Windows release. The 2026 branch adds a new subcategory under System Apps labeled “Artificial Intelligence,” alongside granular entries for specific services like the Windows Copilot runtime and the Recall database engine.
This is not a simple toggle. Removing AI components requires careful selection because some entries have shallow dependencies on other subsystems. NTLite’s compatibility manager highlights these conflicts, but the burden of testing falls on the user. Strip too aggressively, and you might break search indexing or voice typing. The tool’s changelog emphasizes that the feature is experimental and targets only the Windows 11 25H2 build line, which carries a reworked component store that makes this granularity possible.
The process itself is straightforward: load a 25H2 image, navigate to Components, expand the System Apps node, and uncheck the boxes for Copilot in Windows, Windows Recall, AI-powered webcam effects, and the underlying machine learning runtimes. NTLite then rebuilds the image without those packages. The saved ISO installs a version of Windows that, in theory, never phones home to Copilot’s cloud endpoints or stores snapshots of your screen activity.
The AI Backlash Driving Demand
Microsoft’s AI roadmap has been divisive. When Windows Recall was first announced in early 2024, security researchers quickly demonstrated how its unencrypted database could be accessed by malware. The company responded with encryption and opt-in enrollment, but the distrust stuck. Copilot’s integration into the taskbar, Edge, and Microsoft 365 apps also drew criticism from enterprise customers who had not yet completed data governance reviews. Recent surveys among IT professionals show that nearly 40% plan to delay or block Windows 11 25H2 deployments specifically because of AI feature concerns.
NTLite’s new capability taps directly into that anxiety. System administrators managing regulated environments—healthcare, finance, government—now have an officially independent method to quarantine AI components without waiting for Microsoft’s group policies or ADMX templates, which often lag behind releases. Privacy advocates can build a “debloated” image that removes not only the visible apps but also the background services like the Windows Copilot runtime and the semantic indexing used by Recall.
What Gets Removed, Exactly
Based on the component tree exposed in NTLite v2026.04.10936, users can strip out:
- Copilot in Windows – the side-panel chatbot and its shell integration.
- Windows Recall – the timeline and snapshot service.
- Windows Studio Effects – AI-driven background blur, eye contact, and automatic framing.
- ML inference runtimes – the on-device AI engines used for live captions, voice access, and text predictions.
- Cognitive services – APIs that third-party apps can call for AI tasks.
- Telemetry and diagnostics components tied to AI feature usage, which are separate from standard telemetry.
The tool also surfaces a “Recommended cleanup” script that disables scheduled tasks and services that would attempt to re-download removed AI packages from Windows Update. Without this step, some components like Copilot can auto-reinstall because Microsoft classifies them as system apps rather than optional features.
Real-World Impact on Performance and Privacy
Benchmarks run on 25H2 preview builds show that a default installation idles with 200–300 MB of RAM consumed by AI services, plus periodic CPU bursts when Recall takes screenshots or when Copilot preloads its language model. Stripping those components reclaims that memory and eliminates the background I/O, which matters on machines with 8 GB or less of RAM. For older systems, the difference can be the margin between a usable desktop and a swapping mess.
On the privacy front, removing Recall entirely cuts off the data pipeline at the source. Even with encryption, a fully enabled Recall database logs every action—file edits, web pages visited, application usage—in a SQLite database that is, by design, accessible to any administrator process. NTLite’s removal prevents the database from being created in the first place, which is a more complete solution than the Settings toggle that just suspends screen capturing.
Enterprise users will appreciate that image-level removal applies before domain join. That means a fleet of imaged laptops never introduces the risk of a user accidentally enabling Recall and syncing snapshots to a work OneDrive, a scenario that data loss prevention tools are not yet equipped to handle.
Community and Enterprise Reactions
Early chatter on the WindowsForum and NTLite’s own community channels suggests strong uptake. One forum member described the feature as “the only reason I’ll consider 25H2,” while several sysadmins shared scripts that chain NTLite’s command-line interface with WDS and MDT to produce AI-free deployment images automatically. The tool’s developer, known as nuhi, has been responsive to requests for even finer control, hinting at a future “AI services” tab that would allow per-bin removal of DLLs like Windows.AI.MachineLearning.dll.
Microsoft itself has not commented on NTLite’s update, which is unsurprising. The company’s official documentation for 25H2 lists some AI features as removable via DISM or PowerShell, but the list is incomplete and often leaves behind scheduled tasks and registry entries. NTLite’s more aggressive approach exploits the new component store to truly excise those packages, a method that falls outside Microsoft’s support boundaries. Users should be aware that removing AI components may disqualify a system from certain support scenarios, although Windows Update and Security Center functionality remain intact in tests.
Risks and Caveats
Removing AI components is not without side effects. Live captions, which many users rely on for accessibility, depend on the same ML runtimes. Voice typing and the emoji picker’s GIF search may break. The stock Photos app’s AI-based background removal feature will fall back to a grayed-out button. NTLite’s compatibility manager warns about these dependencies, and the tool’s preset editor allows saving a “Safe AI Strip” configuration that balances functionality with privacy by keeping accessibility-related AI engines while discarding Copilot and Recall.
Another risk involves future cumulative updates. Microsoft’s update packages often contain integrity checks for the component store. If an update expects a component to be present and it is not, the update may fail or partially install, leading to a mismatched system. NTLite’s documentation advises updating the base image regularly and testing on a virtual machine before rolling out to production hardware.
How It Compares to Alternatives
Power users have been removing Windows components since the days of vLite for Vista. Tools like MSMG Toolkit and manually editing the Component Store with DISM offer similar outcomes, but NTLite v2026 simplifies the process with a gui and explicit AI categories. Microsoft’s own LTSC editions, which strip out most Store apps and premium features, may eventually offer an AI-light build, but currently 24H2 LTSC still includes Copilot and the AI runtimes by default. Windows 11 IoT Enterprise, another minimalist SKU, also retains the core AI stack for compatibility with edge computing scenarios.
For those unwilling to use third-party tools, Microsoft provides group policies to disable Copilot and Recall, but these are not an image-level removal. A determined user or a misconfigured policy refresh can re-enable them. NTLite’s approach is, in effect, a hardware-level firewall: the bits simply don’t exist on the disk.
Looking Forward
NTLite’s AI removal feature is likely to remain a cat-and-mouse game. Microsoft has shown a pattern of reclassifying components and renaming packages across Insider builds, which forces tool developers to constantly update their detection algorithms. The NTLite team has committed to rapid updates during the 25H2 preview period, urging users to report any new packages that appear in later builds.
The larger trend is unmistakable: as operating systems bake AI deeper into the kernel and shell, the demand for surgical removal tools will only grow. NTLite v2026 represents one of the first mainstream responses to that demand, and it will probably not be the last. Whether Microsoft responds by making its own removal tools more complete or by locking down the component store further is an open question. For now, the power is in the hands of the image editor.