Right-click a photo in Windows 11’s newest Canary build, and a new “AI actions” submenu appears—offering one-click background removal, object erasure, background blur, and Bing Visual Search without ever opening an app. Microsoft shipped Build 27938 to the Canary Channel on September 27, 2023, revealing a deeper integration of generative AI directly into File Explorer’s context menu, a move that signals a broader shift toward surface-level AI assistance across the operating system.
The build debuts four image-centric AI actions: “Remove Background” (which launches Paint’s automatic subject cutout), “Blur Background” and “Erase Objects” (both routed through the Photos app), and “Bing Visual Search” (using the selected image as a web query). These actions work with .jpg, .jpeg, and .png files, covering the most common consumer photo formats. The feature is rolling out gradually via server-side gating, so not every Insider on Build 27938 will see the menu immediately. Alongside the AI additions, Microsoft restored the much-requested clock with seconds in the notification center and added a new Settings page under Privacy & security > Text and image generation that lets users see which apps have used Windows’ local generative AI models and optionally revoke permission.
How AI Actions Work in Practice
The AI actions are implemented as a shell extension: right-clicking a supported image file shows a new “AI actions” entry in the context menu. Selecting an option triggers one of two behaviors. For editing tasks, File Explorer hands off the image to a first-party app—Photos for blur and generative erase, Paint for background removal—with the edit already staged and ready for quick refinement. For visual search, the system invokes Bing Visual Search to find similar images, identify landmarks, extract text, or offer shopping results. In all cases, the goal is to eliminate the friction of manually opening an app, locating the tool, and re-loading the file.
Because the actions lean on Photos and Paint, the exact behavior depends on the versions of those apps installed via the Microsoft Store. Microsoft has not disclosed whether each AI operation runs locally on Copilot+ hardware or offloads to the cloud. The company’s hybrid model means some workloads might execute entirely on-device while others could call Azure endpoints; the blend varies by device capability and Microsoft’s runtime decisions. Until per-action locality guarantees are published, users working with sensitive images should assume that cloud processing is possible for actions like Visual Search and any future document summarization features.
Supported file types at launch are limited to JPEG and PNG. Professional formats such as RAW, PSD, or TIFF are not supported, and community testing indicates that large high-resolution files may experience delays or fall back to slower processing paths. Microsoft has separately confirmed that document-level AI actions—like “Summarize” and “Create an FAQ” for Microsoft 365 files—are on the roadmap, but those will be gated by Copilot/Microsoft 365 licensing and will support a wider set of filetypes including docx, pptx, xlsx, pdf, and txt. Initial availability for document actions is tied to Beta or Microsoft 365 Insider enrollment.
Build Details and Stability Warnings
Build 27938 is explicitly a Canary Channel release, meaning it’s the most experimental and least stable flight. Microsoft’s changelog warns of possible installation rollbacks with errors 0xC1900101-0x20017 or 0xC1900101-0x30017, meaning that some systems will attempt to update only to revert automatically. The build also fixes several bugs from earlier Canary releases: the “Reset this PC” failure in Build 27934, dark-mode color glitches in File Explorer, missing video thumbnails for files with certain EXIF metadata, and a task manager freeze on the performance tab. But it carries new known issues, including screen flickering in browsers and the inability for PIX on Windows to replay GPU captures (a fix is promised in a future PIX release).
Independent reports and forum discussions link this build to ViVeTool feature IDs that can force-enable the AI actions menu, though such tweaks are unofficial and can destabilize the system. The community thread also notes that the build number itself is researcher-reported and not yet confirmed by Microsoft’s Flight Hub, but the official changelog on thewincentral.com matches the company’s typical Insider blog format, lending it credibility.
The Clock Returns—and So Does AI Transparency
One welcome utilitarian addition is the larger clock with seconds in the notification center, enabled via Settings > Time & language > Date & time > “Show time in the Notification Center.” This echoes the beloved Windows 10 feature that power users have been requesting since Windows 11’s launch. More strategically significant for enterprise IT is the new “Text and image generation” page in Privacy & security settings. It lists third-party applications that recently used Windows-provided generative AI models—a first step toward auditing and controlling which apps can tap into local AI capabilities. This page is currently read-only but establishes a foundation for future group policy and MDM controls.
Community Perspectives: Convenience vs. Control
The response from Windows enthusiasts has been mixed but largely interested. On forums, users praise the reduced context-switching: removing a background from a product photo or blurring a distracting element without launching a heavy editor can save dozens of seconds per task, which compounds rapidly for content creators and marketers. One power user noted that while the edits are basic, they cover the most common micro-tasks that previously justified the price of third-party utilities. Others expressed unease about data privacy, particularly because the right-click menu is so discoverable; a casual user might unknowingly upload a sensitive screenshot to Bing Visual Search.
Enterprise administrators flagged several governance gaps. The new settings page shows app activity but doesn’t provide per-action toggles, audit logs, or network egress controls. Without explicit guidance on which actions use cloud endpoints, organizations in regulated industries will be hesitant to allow the feature on devices handling confidential data. The licensing complexity for document actions—requiring Copilot or Microsoft 365 subscriptions—adds another layer of management overhead, as mixed license environments could lead to inconsistent experiences for users.
Privacy and Security: What You Need to Know
Microsoft’s hybrid execution model is the key privacy concern. While the “Text and image generation” page provides visibility into local model usage, it does not cover cloud-based processing. Visual search, by its nature, must send the image to Bing’s servers. Blur and erase, when routed through Photos, might run locally on capable hardware, but the app’s terms of service could still allow cloud processing for certain generative features. Microsoft has not published a decision tree clarifying these paths. For now, administrators should treat all AI actions as potentially cloud-bound and implement data loss prevention (DLP) policies accordingly.
The context menu’s convenience also raises the risk of accidental uploads. A single errant right-click could send a confidential document or internal screenshot to a search engine. User education is essential during Insider testing, and enterprises should consider disabling the feature via group policy once administrative templates become available. Currently, there are no GPO or MDM options to block AI actions; the settings page is merely informational.
How to Try AI Actions Today
Only Insiders on the Canary or Dev channels (after eventual promotion) will see Build 27938. Even then, the AI actions feature is server-gated, so it might not appear immediately. For those determined to test:
1. Enroll a non-production device or VM in the Canary or Dev channel.
2. Update to the latest build via Windows Update.
3. Right-click a .jpg, .jpeg, or .png file in File Explorer and look for “AI actions” in the context menu.
4. If the entry doesn’t appear, server-side gate hasn’t activated for your machine. Advanced users may use ViVeTool to force-enable feature IDs reported by the community (e.g., ID 50032563 and others), but this is unsupported and risky.
5. Test actions only on non-sensitive images until you confirm where processing occurs.
6. Keep Photos and Paint apps updated through the Microsoft Store.
Remember that Canary builds can roll back or cause driver regressions. Do not install them on daily-driver hardware without a full backup.
Strengths and Weaknesses of the Feature
Strengths:
- Productivity gains: Instant edits for common tasks eliminate app-switching.
- Leverages existing apps: No new toolkit to learn; uses familiar Paint and Photos interfaces.
- Governance visibility: The new AI activity page is a promising start for enterprise management.
Weaknesses:
- Privacy opacity: Unclear which operations are local vs. cloud.
- Context menu clutter: Additional entries may frustrate power users who prefer a lean right-click menu; no built-in customization.
- Licensing fragmentation: Document actions will require Copilot subscriptions, creating a two-tier experience.
- Stability risks: Canary builds are inherently unstable; this one specifically warns of rollback failures.
What This Reveals About Windows’ AI Roadmap
Surface-level AI actions in File Explorer are part of a deliberate strategy to weave intelligence directly into the shell, reducing reliance on a single Copilot app. By embedding editing and search into the context menu, Microsoft is betting that users will adopt AI more readily if it appears where they already work. This “invisible AI” approach mirrors the company’s broader push toward ambient computing, but it also raises the bar for transparency and administrative control. The incremental rollout—Canary first, then Dev, then Beta, and finally general availability—gives Microsoft room to gather telemetry and refine based on Insider feedback.
The real inflection point will come when per-action data locality guarantees are published, enterprise management hooks are added, and licensing terms are clarified for consumers. For now, Insiders and IT administrators should consider this a preview of a direction, not a finished feature.
The Bottom Line
AI actions in Windows 11’s File Explorer are a pragmatic, high-leverage experiment. They put time-saving image edits and visual search one right-click away, lowering the skill barrier for casual creators while offering a taste of the OS’s future. But the feature’s success hinges on Microsoft’s ability to address privacy opacity, provide governance controls, and stabilize the build. If you’re curious, test on a spare device, keep sensitive data offline, and file feedback via the Feedback Hub. For production environments, wait until these capabilities reach Beta or Release Preview with clear documentation and policy tools. Microsoft is actively listening and iterating—your feedback during these early flights will shape the final experience.