A new right-click context menu in Windows 11 File Explorer is about to make casual image editing as simple as a single click. Microsoft has begun rolling out an “AI Actions” submenu in the latest Canary channel Insider previews, allowing users to blur backgrounds, erase objects, remove backgrounds, or run a visual search without ever launching a separate app. Community reports pinpoint the feature’s appearance in Build 27938, though Microsoft’s staged rollout means not every Insider device will see it yet.

This move signals a deeper integration of generative AI into the Windows shell, turning the file manager into a launchpad for micro-workflows that previously required the Photos or Paint apps. In this article, we break down exactly what’s included, how the technology works, the privacy controls now being tested, and what IT administrators need to know before these features reach production.

What’s Included in the AI Actions Preview

The AI Actions submenu appears when you right-click a supported image—currently .jpg, .jpeg, and .png files—inside File Explorer. It groups four distinct capabilities:

  • Bing Visual Search – Uses the selected image as a search query to find visually similar pictures, identify landmarks, plants, or products, and locate web pages that contain the same image. Essentially, it’s a reverse image search built into the shell.
  • Blur Background – Opens the Photos app with automated subject/background separation already applied. A single click blurs everything behind the subject, and you can adjust intensity or refine the selection with a brush tool.
  • Erase Objects (Generative Erase) – Opensa the Photos app’s generative erase flow. Select an object in the image, and the model fills the removed area with content-aware blending. Microsoft markets this as a quick way to eliminate photobombers or distractions.
  • Remove Background – Routes the image directly to Paint’s automatic background-removal pipeline, producing a subject cutout in one step. The result is a transparent-background PNG, ideal for pasting into presentations or collages.

All four actions are, in essence, orchestration layers. They hand off the image to a first-party app (Photos or Paint) with the edit pre-staged, or invoke a platform API for cloud‑based services like Bing Visual Search. This design reuses mature tooling rather than reinventing wheels, but it also means the output quality is directly tied to whatever model powers the underlying app.

Microsoft has publicly stated that this is an initial, image‑focused set. Future rollouts, likely tied to Microsoft 365 Copilot licensing, may add document‑level actions like summarization or FAQ generation for Word and PDF files right from the File Explorer context menu.

How It Works Under the Hood

AI Actions are implemented as shell hooks—extensions to the File Explorer context menu. Technically, they fall into two categories:

  1. App‑staged workflows – The shell Opens a target app (Photos/Paint) with a URI that triggers a specific edit mode and passes the file path. The user then sees the prepared edit and can tweak or confirm it.
  2. API‑mediated operations – Some actions, like Bing Visual Search, invoke a platform service that may run either locally (often on Copilot+ PCs with an NPU) or in the cloud. Microsoft has not published a detailed per‑action locality matrix, so whether an operation processes entirely on‑device or sends data to Microsoft servers depends on hardware, software version, user settings, and possibly licensing.

This hybrid approach is both a strength and a source of ambiguity. On a Copilot‑enabled machine with a neural processing unit, tasks like background removal could theoretically run entirely locally, offering low latency and stronger privacy. On standard hardware, the same click might route the image through a cloud endpoint. Microsoft’s public documentation so far does not clarify which actions always stay local. Without that guarantee, organizations handling regulated data must assume that egress is possible.

A New Transparency Dashboard

Alongside the AI Actions menu, Microsoft is testing a companion Settings page at Settings > Privacy & security > Text & image generation. This page lists third‑party applications that have recently used Windows‑provided generative AI models. For each app, a toggle lets users allow or block that app’s access to the AI platform.

This is a notable first step toward OS‑level governance of AI features. It gives end users visibility into which apps are leveraging the AI stack, and puts a coarse-grained kill switch in the hands of anyone concerned about data leakage. However, it does not yet replace enterprise tools like Microsoft Intune or Group Policy for managing AI model access across fleets. Admins will need to combine this setting with network controls and conditional access policies until more granular, centrally manageable controls are released.

Productivity Gains and UX Benefits

The shift to shell‑integrated AI edits delivers several tangible productivity improvements:

  • Fewer context switches – Small editing jobs no longer require opening a full editor, navigating menus, and saving a copy. One click gets the job done.
  • Better discoverability – Users who never open advanced photo tools can still perform common tasks like removing a background or blurring a face. The right‑click menu is universally familiar, dramatically lowering the learning curve.
  • Faster micro‑workflows – Tasks like removing an object or background are often minute‑scale chores. Summing up those saved seconds across a day of social‑media posts or presentation building can add up to meaningful time savings.
  • Unified entry point – By aggregating visual search and quick edits under one menu, Microsoft creates a single place to get answers about an image or clean it up, without hunting through separate apps.

Early hands‑on reviews from Insiders confirm that the experience feels snappy when it works, particularly for the Blur Background and Remove Background actions that rely on local pre‑processing. However, the exact speed and quality vary with hardware, and some testers note occasional artifacts in generative erase results—mismatched textures or blurry infill that may not pass muster for professional use.

Risks, Limitations, and What’s Missing

Despite the promise, there are several important caveats to keep in mind, especially for power users and IT departments.

Channel instability and server gating – Features in the Canary, Dev, and Beta channels are not final. They can appear and disappear via A/B testing, feature‑flag toggles, and server‑side changes. Even on the same build, two machines may see different menus. Don’t rely on AI Actions for critical workflows during the preview phase.

Limited file format support – Only JPEG and PNG are supported at launch. RAW, PSD, TIFF, and other pro formats are excluded, so photographers and designers will still need full editors for raw processing or layered files.

Model artifacts – Generative erase and background removal can produce blending errors, especially around complex edges like hair or transparent objects. Microsoft’s models are improving, but you should inspect results carefully before using them in client work or public‑facing assets.

Ambiguous data flow – As noted, the lack of a clear per‑action locality matrix complicates privacy assessments. If your organization handles sensitive images, assume that an action might send data to Microsoft’s servers unless you configure network blocks or disable the feature entirely.

Hardware and licensing gates – Some on‑device execution is optimized for Copilot+ hardware with an NPU, and future document‑level actions may be locked behind Microsoft 365 Copilot subscriptions. This creates a two‑tier experience where the best performance and widest feature set require specific hardware and ongoing licensing costs.

Potential for misuse – Tools that make it trivially easy to remove objects or backgrounds also lower the barrier for image manipulation. In contexts where provenance matters—legal, journalism, scientific imaging—users must be trained to preserve originals and verify edits.

Enterprise and IT Implications

For IT administrators, the arrival of AI Actions in File Explorer is more than a curiosity. It represents a new OS feature that can affect data egress, support desk volume, and security posture. Here’s a step‑by‑step readiness checklist:

  1. Test in a controlled environment – Deploy the Canary build on a representative VM or spare PC. Evaluate UI behavior, output quality, and any network traffic generated by each action.
  2. Review the new Settings page – Visit Privacy & security > Text & image generation. Understand which apps are using Windows AI models and use per‑app toggles to block access on highly sensitive machines.
  3. Enforce network controls – Until Microsoft publishes a guaranteed on‑device processing list, treat File Explorer AI Actions as potential egress vectors. Use firewall rules, conditional access, or MDM policies to limit cloud access for the relevant MS endpoints.
  4. Update security baselines – Incorporate the new AI capabilities into your threat model and incident response playbooks. Account for image‑based data leaks, metadata exposure, and accidental sharing of unverified edits.
  5. Educate end users – Teach staff the limitations (artifacts, file type support) and the importance of keeping originals. Set a clear policy on when generative edits are appropriate versus when manual editing is required.

Microsoft is likely to release more granular management controls (Intune CSPs, Group Policy) once these features graduate from Canary to broader channels. Watch for updated ADMX templates and security baselines in future Windows releases.

How to Start Testing Today

If you’re eager to try AI Actions yourself, follow these steps:

  • Join the Windows Insider Program – Go to Settings > Windows Update > Windows Insider Program and select the Canary or Dev channel. The feature is rolling out in stages, so you may not see it immediately even after updating.
  • Install the latest build – Once enrolled, check for updates and install a build that includes the AI Actions submenu. Community reports point to Build 27938, but your experience may differ.
  • Try it out – Right‑click any JPEG or PNG file in File Explorer. If the feature is enabled, you’ll see an “AI Actions” submenu. Click an action and observe how the workflow unfolds.
  • Check transparency – After using an action, navigate to Privacy & security > Text & image generation to see which apps recently accessed Windows AI models.

A word of caution: some guides recommend using third‑party tools like ViveTool to force‑enable hidden flags if the menu doesn’t appear. That approach is unsupported and can cause instability. Only use such tools in isolated test VMs, never on production or corporate devices.

The Road Ahead

Microsoft’s File Explorer experiment is a landmark moment in the company’s AI integration strategy. It transforms a decades‑old file manager into a canvas for generative AI, reducing friction for everyday tasks while surfacing capabilities that many users never knew existed.

Looking forward, several developments are likely:

  • Broader format support – Adding RAW, PSD, and TIFF handling would require non‑destructive editing pipelines and metadata preservation. This is a natural next step but a technically heavy lift.
  • Document‑level actions – Summarization, FAQ generation, and translation for Office files, potentially locked behind Microsoft 365 Copilot subscriptions.
  • Enterprise controls – Expect MDM/Group Policy settings that allow admins to disable AI Actions entirely, block cloud processing, or audit usage across the organization.
  • Clarity on locality – Microsoft must eventually publish a clear document stating which AI Actions run fully on‑device across different hardware configurations, so that regulated industries can make informed decisions.

The balance between convenience and control will determine whether AI Actions become a universally welcome productivity boost or a feature that enterprises feel compelled to disable until governance catches up. For now, it’s an exciting preview that gives Windows enthusiasts a tangible taste of an AI‑native OS.