Windows 11 26220.7262 Insider Preview: HD Voices, AI Tools, CFR Rollout

Windows Insiders on the Dev and Beta channels are receiving Windows 11 Insider Preview Build 26220.7262 (KB5070303), a 25H2-based enablement-package release that smooths the path for several incremental accessibility, AI, and usability improvements while keeping a careful, controlled rollout through toggles and CFR (Controlled Feature Rollout) mechanisms.

Background / Overview​

Microsoft is distributing Build 26220.7262 to both the Dev and Beta Channels as part of its ongoing 25H2 enablement-package stream. For a limited time, Insiders in the Dev Channel can switch to Beta while these 25H2-based builds remain identical across the two channels. Once the Dev Channel advances to a higher build, the window to switch will close and the Dev Channel may again diverge into less stable, experimental builds.
This release continues Microsoft’s approach of staged delivery: some features are rolled out gradually to Insiders who opt in to receive the very latest updates via Settings > Windows Update, while other improvements are expanded to everyone in the channels over time. That model allows Microsoft to gather telemetry and Feedback Hub input before wider exposure.
Key themes in this build include:

• Accessibility upgrades (HD voices for Narrator and Magnifier; structured math reading in Narrator)
• New AI-related settings and temporary reconfigurations of File Explorer AI capabilities
• UX refinements (Click to Do context menu, teaching tips) and hardware feedback (haptic pens)
• Bug fixes and a small set of known issues for Insiders to watch

---

What’s new in Build 26220.7262​

HD voices for Narrator and Magnifier (on-device, expressive speech)​

Microsoft introduced high-definition (HD) voices accessible in both Narrator and Magnifier. These are powered by the latest on-device text-to-speech models and aim to produce more natural, expressive speech with smoother prosody and pacing.

• The initial offering is focused on English (United States) and provides at least two persona options: Andrew (male) and Ava (female).
• HD voices are delivered as downloadable options inside Settings → Accessibility → Narrator (and the same flow in Magnifier): tap “Add a natural voice,” download your preferred HD voice, and switch or adjust playback speed at any time.
• The key technical detail is that HD voices use newer on-device TTS models designed to reduce listening fatigue during long sessions and to read content with more natural intonation.

Benefits:

• Clearer audio and more lifelike cadence when Narrator or Magnifier reads documents and on-screen content.
• Lower cognitive load for users who rely on screen readers for extended reading or complex content.

Practical notes and caveats:

• HD voices are currently localized only to en-US in this early rollout; additional locales are expected over time.
• Because these voices use more advanced models, they may have a larger footprint on disk once downloaded. Users should manage storage on devices with constrained space.
• The experience is opt-in per-device (download required) so enterprise IT can limit deployments if desired.

Narrator now supports structured math reading (first phase)​

This build introduces the first phase of Math reading in Narrator, enabling more natural and structured playback of equations and scientific notation in supported scenarios — notably Microsoft 365 apps such as Word.
How it works:

• With Narrator enabled (Win + Ctrl + Enter), standard Narrator reading commands will read equations in a way that reflects mathematical structure rather than a flat stream of characters.
• The intent is to make math accessible to blind and low-vision users in educational and professional contexts.

Limitations and expected progress:

• This is an initial phase and is scoped to certain app scenarios (Word and Microsoft 365 documents are explicitly called out).
• Microsoft plans to broaden coverage over time and refine the reading commands and structure parsing.
• Users who rely on precise math rendering should verify behavior on sample documents and report gaps through Feedback Hub to accelerate improvements.

AI Components: “Experimental agentic features” toggle​

A new Settings toggle appears under System → AI Components labeled “Experimental agentic features.” This setting controls whether Windows agents can use new agentic capabilities—functions that enable agents to perform multi-step or proactive tasks on behalf of users.
Why this matters:

• Agentic features move beyond single-query responses toward agents that can sequence actions, make decisions, and potentially interact with system functions or third-party services.
• The toggle gives Insiders control over exposure to early agentic behaviors while Microsoft tests the UX, safety, and telemetry aspects.

Security and privacy considerations:

• Agentic features are experimental and may surface new telemetry and decision-making behaviors. Insiders should treat them as experimental and enable them only on test devices or where the privacy implications are acceptable.
• Organizations should be cautious about enabling experimental agentic options on production endpoints until guidance and enterprise controls are more mature.

Click to Do: streamlined context menu and auto-pop for images/tables​

The Click to Do context menu is receiving a design refresh aimed at making common actions easier to find. The new design surfaces frequently used actions such as Copy, Save, Share, and Open more prominently.
Notable behaviors:

• When a large image or table appears on the screen, Click to Do may automatically pop up with suggested actions, shortening the time from selection to execution.
• This is part of Microsoft’s wider push to surface AI-driven actions in context where they’re most useful.

Temporary change:

• Microsoft also reported turning off Image Object Selection temporarily for Dev and Beta Insiders. This indicates a short-term rollback while the company adjusts the functionality or fixes underlying issues.

File Explorer AI actions, CFR reconfiguration and tabs pause​

File Explorer’s AI actions (including Copilot summary actions and image actions) are undergoing a CFR reconfiguration. Insiders may notice a temporary reduction in image-based actions and Copilot summaries as the system is relaunched and gradually re-enabled.
Additionally, the rollout of a change that would open newly launched folders in tabs (instead of separate windows) has been paused due to reliability issues. Microsoft is temporarily stopping that part of the roll-out and will reintroduce it once fixed.
Implications:

• Insiders testing File Explorer’s AI integrations may see fluctuations in availability while Microsoft refines backend configuration and rollout logic.
• These temporary pauses are typical of CFR-driven rollouts where feature subsets are iterated upon before broader exposure.

Teaching tips, haptic pen support, and UX polish​

The release includes several smaller but meaningful UX changes:

• Teaching tips and the tutorial introduction screen have been refreshed with a new “Launch Tutorial” button and clearer onboarding instructions.
• Support for haptic feedback with pens that expose the capability: tactile responses may occur during interactions like hovering over close buttons or snapping windows.

Real-world impact:

• Haptic responses can improve discoverability and provide a tactile layer to UI affordances, but the feature is hardware-dependent (requires supported pens and drivers).
• Teaching tip improvements are helpful for new users and should improve the onboarding experience for new or lesser-known UI features.

---

Fixes included in this build (selected)​

The build brings fixes for several issues that had degraded usability for some Insiders:

• Fixed a problem where mouse and keyboard input ceased to work for certain Insiders in WinRE / Advanced Startup environments.
• Resolved a Taskbar-related bug that incorrectly told some Insiders their camera was ineligible for Recall features.
• Addressed an issue where Task Manager’s process did not stop correctly after closing, which in some cases caused Task Manager to open unexpectedly on boot.
• Restored Virtual Workspaces options that were not functioning correctly under Settings → System → Advanced.

These fixes are targeted at improving stability and reliability in areas that matter to power users, IT pros, and developers who test early builds.

---

Known issues you should know about​

Microsoft lists several known issues for Insiders running this build:

• Taskbar and System Tray: Some Insiders may find the Start menu does not open via click (it still opens with the Windows key). The behavior may be linked to the notification center (WIN + N).
• System Tray Visibility: Certain apps may not appear in the system tray even when they should be visible.
• File Explorer copy dialog (dark mode): Copy progress may flash when toggling details; a missing scrollbar and footer may show a white block when text scaling is applied.
• .NET Framework / Visual Studio (ARM64): On ARM64 PCs, Visual Studio or apps relying on .NET Framework may experience crashes. Microsoft’s recommended mitigation is to check Windows Update and install the latest .NET Framework fixes as they become available.

Practical steps if you encounter these issues:

1. Use the Feedback Hub to report reproducible problems (WIN + F). Include repro steps and diagnostics when possible.
2. Check Windows Update for .NET Framework patches if you run ARM64 workloads and Visual Studio.
3. Avoid enabling experimental toggles on production devices; use a test system for bleeding-edge features.

---

Cross-check: technical details and verification​

Several technical details in this release align with independent public sources and product documentation:

• Azure’s HD TTS models and the “DragonHD” family are published in Microsoft’s Azure AI Speech documentation, showing en-US voices such as Andrew and Ava are part of the available persona set.
• Microsoft’s staged rollout model and Controlled Feature Rollout (CFR) behavior are consistent with past Insider distribution patterns and the presence of opt-in toggles for “get the latest updates” in Settings → Windows Update.
• The concept of agentic features (agents performing multi-step actions) and a dedicated toggle for experimentation follows Microsoft’s recent trajectory toward integrating more agent-based AI features and providing user-level controls.

Where claims could change or remain fluid:

• Availability of HD voices across regions and additional locales depends on Azure region rollout and Microsoft’s internal schedule; Insiders should expect expansion over time.
• Agentic features are explicitly experimental; their behavior, privacy model, and enterprise controls may evolve before any broad public release.

---

Risks, privacy, and security analysis​

The build introduces several capabilities that raise important security and privacy questions—particularly around AI-driven and agentic features.
Key risk areas:

• Agentic features: Allowing agents to act autonomously can expand attack surface and increase the chance of inadvertent data exposure if an agent acts on third-party integrations. Experimental toggles help, but enterprises should not enable agentic features on critical assets without policy controls.
• On-device vs. cloud AI: HD voices are presented as on-device models, which reduces cloud dependency and improves privacy, but many AI interactions in Windows still rely on cloud services. Users and organizations should review telemetry, network policies, and local caching behavior.
• File Explorer AI: Summaries and image actions may process content that organizations consider sensitive. Controlled rollouts are appropriate, but admins should audit what is transmitted to AI services and whether policy controls exist to prevent exfiltration.
• Beta/Dev channel mixing: The temporary alignment of Dev and Beta builds offers the convenience of switching channels but increases the chance that Insiders expecting Beta-level stability will see more experimental behaviors if they switch back and forth. Users should test on nonessential devices.

Recommended mitigation steps:

• Use test hardware for Insider builds; don't enable experimental toggles on workstations handling confidential data.
• Review Windows privacy and diagnostic settings, and configure network controls where needed to limit cloud calls for AI features.
• Monitor for updates to Microsoft’s enterprise guidance for agentic features or AI Components settings.

---

Guidance for Insiders: should you switch channels or enable the toggle?​

For Insiders deciding whether to switch from Dev to Beta while the builds match:

• If you prioritize stability and fewer surprises, stay in Beta or wait to switch only after reviewing community feedback and known-issue lists.
• If you want earlier access to incremental UI tweaks and AI experiments and are comfortable troubleshooting, remaining in Dev while the window is open is acceptable—but back up your data.

Regarding the “get the latest updates as they are available” toggle:

• Turn the toggle on if you want to receive gradual CFR-based feature rollouts sooner and can tolerate intermittent issues.
• Keep it off if you prefer features to arrive fully matured and broadly validated.

Operational best practices:

1. Always image or snapshot your test machine before major Insider updates.
2. Use Feedback Hub deliberately—file reproducible steps and attach diagnostics.
3. For enterprise testers, coordinate with IT to ensure Insider features don’t collide with compliance or endpoint protection policies.

---

What to watch next​

Insiders should monitor a few areas as Microsoft continues to iterate:

• Broader availability of HD voices (more languages and region support) and any performance/space trade-offs on smaller devices.
• Expansion of math-reading support beyond Word and Microsoft 365, including web and PDF scenarios.
• How the “Experimental agentic features” toggle evolves: documentation, enterprise controls, and telemetry options.
• File Explorer AI reconfiguration: whether image actions and Copilot summaries return in a modified form and when folder-tabs behavior is reintroduced.
• Fixes for the Start menu, system tray, and File Explorer copy dialog dark mode issues.

---

Final assessment​

Build 26220.7262 is a measured update that blends meaningful accessibility advances with continued AI experimentation. The HD voices and Math reading improvements are welcome for users who rely on Narrator and Magnifier, and the explicit Experimental agentic features toggle reflects Microsoft’s attempt to give users control over emergent agentic behaviors.
At the same time, the build surfaces typical trade-offs: temporary rollbacks (Image Object select, File Explorer tabs), a small set of stability concerns, and features gated behind CFR and opt-in toggles. Those trade-offs are by design and reflect a staged, feedback-driven approach that suits Insider testing, but they underscore why Insiders should run these builds on noncritical devices and report issues through Feedback Hub.
For enthusiasts and IT pros who want to help shape Windows, this build provides the right mix of accessibility wins and AI experimentation to test and critique. For users who need rock-solid stability, wait for wider rollouts and enterprise guidance on agentic features before enabling experimental toggles in production environments.
Overall, Build 26220.7262 continues Windows’ push toward richer accessibility, contextual AI, and agentic experimentation—while reminding Insiders that careful testing and conservative deployment practices remain essential during the 25H2 enablement cycle.

---

Source: Microsoft - Windows Insiders Blog Announcing Windows 11 Insider Preview Build 26220.7262 (Dev & Beta Channels)

 

[ChatGPT]

Microsoft is rolling out a targeted Insider preview for Windows 11 25H2 today with Build 26220.7262 (KB5070303), a compact but consequential update that layers accessibility upgrades, UI refinements and — most notably — an explicit toggle for experimental agentic features that lets Windows host AI “agents” capable of acting on your behalf under tightly scoped conditions. This build is being offered to both the Dev and Beta channels as part of Microsoft’s enablement-package approach to 25H2 and continues the staged, telemetry-driven rollout model that gates some features by hardware, entitlement and controlled feature flags.

Background / Overview​

Windows 11 version 25H2 is being shipped to Insiders largely as an enablement package layered on the current servicing branch. That means many new features are already present in shipped binaries and are being turned on — or held back — using Controlled Feature Rollout (CFR) logic. The 26220.xxxx family represents the 25H2 preview line in Dev and (temporarily) Beta, and Microsoft continues to deliver frequent small cumulative packages (KBs) to tune behaviors and test staged AI surfaces like Click to Do, File Explorer AI actions, and Copilot integrations. This particular build is not a radical UI rework; it’s a preview-quality update that focuses on:

• Accessibility improvements (high‑definition voices in Narrator and Magnifier; initial Math reading support in Narrator).
• An “Experimental agentic features” toggle surfaced in Settings that enables agentic runtime primitives.
• Click to Do refinements, File Explorer/CFR reconfiguration, and minor UX fixes (teaching tips, haptic pen behavior).
• Bug fixes for issues such as Task Manager persistence and WinRE input failures, and a small set of known issues to watch for.

---

What’s new in this build (practical summary)​

Accessibility: HD voices and Math in Narrator​

Narrator and Magnifier now support high‑definition (HD) voices in English (United States) — labeled in this flight as two downloadable voices, Andrew (male) and Ava (female). These voices use Microsoft’s on‑device, neural text‑to‑speech models to deliver smoother intonation, adaptive pacing and less listening fatigue for longer reading sessions. The new voices are available via Settings → Accessibility → Narrator (or Magnifier) → “Add a natural voice.” In parallel, Microsoft is rolling out the first phase of math reading in Narrator: equations authored in Microsoft 365 (for example Word) can be read back in a more structured, semantically meaningful way that aims to reflect mathematical grouping and natural speech patterns. This is an accessibility-first improvement that directly benefits blind and low‑vision STEM readers. To try it: enable Narrator (Win + Ctrl + Enter) and open a document with equations. Feedback should be filed through Feedback Hub under Accessibility > Narrator. Why this matters: the combination of HD voices and structured math rendering is a meaningful accessibility leap — it reduces friction for long-form listening and educational content and brings on‑device generative speech quality to built‑in Windows assistive tech. Independent Azure TTS advances (and Microsoft’s own on‑device voice model work) make these experiences possible without forcing cloud round trips in many scenarios.

Experimental agentic features: the Settings toggle that changes the trust model​

This build introduces a new Settings control: System → AI components → Agent tools → Experimental agentic features. Turning this toggle on is the explicit opt‑in that allows Windows to provision the runtime support required for agentic workflows — agents that can perform multi‑step tasks on a user’s behalf (open apps, manipulate files, click UI elements, chain operations). Microsoft frames this as experimental and disabled by default, and the company has published guidance and a security posture describing agent accounts, agent workspaces, scoped permissions and signing requirements. Key design primitives Microsoft is using for agentic safety:

• Agent workspace: a contained, parallel desktop session where agents execute, visible and interruptible to the user.
• Agent accounts: agents run under separate standard Windows accounts to enable distinct ACLs and auditing.
• Scoped permissions: initial access limited to “known folders” (Documents, Desktop, Downloads, Pictures) unless the user grants more.
• Signing and revocation: agents must be digitally signed so provenance and operational revocation are practical.

This is not theoretical: Microsoft has already previewed Copilot Actions (web and device) in Copilot Labs, and the Settings toggle simply exposes the on‑device runtime gating for those same, more capable experiences. The company’s security blog discusses these primitives and reiterates that the experience is staged, opt‑in, and subject to ongoing refinement.

Click to Do, File Explorer, and other incremental UX changes​

• The Click to Do context menu is being streamlined to make common actions (Copy, Save, Share, Open) easier to find. The context menu may now automatically surface when a large image or table is detected.
• Image Object Select is temporarily turned off for Dev and Beta while Microsoft refines the implementation. Likewise, AI Actions in File Explorer and Copilot summary actions are being reconfigured under CFR and may be temporarily unavailable for some Insiders.
• Pens that support haptic feedback will now produce tactile responses for certain system UI interactions, adding a small physical layer of affordance to pen use.

Fixes and known issues​

Notable fixes in this build include:

• Resolved a WinRE bug that rendered mouse and keyboard unusable for some Insiders during Advanced Startup.
• Fixed a Task Manager hang where the process didn’t stop correctly after closing, which could lead to Task Manager re‑opening on boot.
• Repaired Virtual Workspaces options in Settings > System > Advanced.
  Known issues to watch for: Start menu failing to open on click (Win key still works), apps not appearing in the system tray, copy dialog display glitches in dark mode, and a new advisory that ARM64 developers may see crashes in Visual Studio or .NET Framework apps until the latest .NET updates are applied.

---

Deep analysis: what Microsoft is testing — and why it matters​

From advice to action: agentic capabilities reshape the desktop​

Historically the OS provided a set of APIs and user controls; Copilot-era development pushes Windows toward being an execution layer for agentic automation. The experimental toggle in this build represents a deliberate product step: Microsoft isn’t quietly enabling agentic primitives — it’s surface‑exposing the permission gate and the runtime model so Insiders can opt in and provide real‑world feedback. When enabled, agents may do more than reply with text — they can run multi‑step workflows, touch local files and interact with installed apps. That capability compresses productivity flows (one instruction instead of many manual steps) at the cost of a materially different trust model. Why this deserves operational scrutiny:

• Agents can cause side effects (move/delete files, send messages) that are irreversible if workflows are incorrect.
• UI automation can be brittle: agents that rely on screen‑level recognition can misclick after app updates or localization changes.
• New attack surfaces appear (prompt and cross‑prompt injection, credential scope creep) that existing endpoint tooling may not detect if the agent uses UI automation rather than sanctioned APIs.

Multiple independent outlets and Microsoft’s own security blog emphasize the same pattern: opt‑in by default, agent accounts, visible workspaces and signing are sound starting points, but they are not a complete solution for enterprise deployment at scale. The missing pieces include mature DLP/Entra/Intune controls, fine‑grained policy for regulated data, robust rollback/transaction semantics and a vetted supply chain process for third‑party agents.

The Copilot+ hardware story and local vs cloud execution​

Microsoft’s hybrid approach separates small, latency-sensitive on‑device models from heavier cloud reasoning. Copilot+ PCs (devices with a qualifying NPU, often described around a 40+ TOPS practical baseline) will unlock lower‑latency, more private on‑device capabilities. On standard hardware, agentic operations will fall back to cloud services for heavier reasoning. This creates a two‑tier experience where privacy and speed are tied to silicon. That puts procurement questions back on the table: does your organization want to standardize on Copilot+ devices to keep sensitive workloads on device?

Accessibility and AI: substantial win, low risk​

The Narrator and Magnifier updates are straightforward quality-of-life and accessibility wins. On‑device HD voices and math rendering reduce reliance on cloud services for everyday assistive scenarios, and they’re controlled within Accessibility settings. These changes have clear positive user value and far fewer systemic risks than agentic automations — though Microsoft will want to validate localization, tuning and model footprints across low‑end devices.

---

Security and privacy: recommended guardrails and operator checklist​

For IT teams evaluating this build (or the agentic features when broader rollout begins), treat agentic features as experimental infrastructure that changes the endpoint threat model. Recommended steps:

1. Test in a controlled pilot ring (VMs or isolated devices) before enabling agent features for any user group.
2. Keep the default toggle off across production fleets until Intune/Entra policy controls and DLP integrations are available.
3. If enabling for testing:
   • Restrict agent access to non‑sensitive folders first (Documents, Desktop, Downloads, Pictures only).
   • Require multi‑party approval for any agent that will act on regulated datasets.
4. Require signed and vetted agents only; implement a certificate/publisher allowlist and revocation plan.
5. Ensure robust backups and clear rollback procedures (file snapshots, versioning) before allowing agents to perform destructive operations.
6. Monitor agent telemetry and session logs closely; ensure log retention and privacy settings conform to local regulations.
7. Ensure ARM64 .NET updates are deployed to avoid Visual Studio/.NET crashes reported for some ARM64 setups.

These steps align with Microsoft’s own guidance to keep agentic experiences opt‑in, observable and revocable during preview. Enterprises should press Microsoft and their device vendors for explicit management hooks (MDM policies, allow/deny lists, audit tooling) before enabling this capability broadly.

---

Practical guidance for Windows enthusiasts and Insiders​

• If you are an enthusiast who wants to experiment: join Copilot Labs and enable Experimental agentic features only on a non‑critical test device. Monitor behavior, file access prompts and agent workspaces closely. Use the visible agent workspace’s step‑by‑step progress to learn how agents behave before granting broader permissions.
• If you rely on Narrator or Magnifier: test the new HD voices and math reading on documents relevant to your workflows; the new voices are downloadable and switchable in Accessibility settings. Expect to give feedback via Feedback Hub if you encounter pronunciation or math parsing edge cases.
• If you manage developer machines (especially ARM64): apply the latest .NET Framework updates as recommended in the build notes to avoid reported crashes with Visual Studio and .NET‑dependent apps.

---

Strengths and potential risks — a candid appraisal​

Strengths​

• Pragmatic rollout model: Microsoft’s enablement package + CFR model reduces blast radius while allowing rapid iteration across thousands of devices. This makes it possible to trial ambitious features like agentic automation without a single mass rollout.
• Built‑in guardrails: agent accounts, agent workspaces, signing and scoped permissions are constructive design choices that materially reduce the risk of silent, uncontrolled automation. These controls make auditing and revocation technically feasible.
• Accessibility wins: HD voices and structured math reading address long-standing assistive tech gaps and are low risk while offering real user benefit.

Risks and open questions​

• New attack surfaces: agents that click and type create a different class of risk (prompt injection, cross‑prompt manipulation, UI fragility) that endpoint detection and DLP tools may not yet detect. Microsoft acknowledges this but enterprise controls are still evolving.
• Supply chain and vetting: requiring a digital signature is necessary but not sufficient; signing pipelines, issuance controls and revocation responsiveness will determine how quickly a compromised agent can be blocked. Details on vetting third‑party agents remain sparse.
• User comprehension and consent fatigue: frequent or opaque permission prompts risk becoming routine clicks. Usability studies and careful prompt design are needed to ensure consent remains meaningful.
• Rollback semantics: transactional undo for multi‑step agent operations (atomic rollbacks, snapshotting) is not yet guaranteed; that increases the severity of mistakes when agents operate on large data sets.

---

What to watch next​

1. Microsoft’s continued documentation and private previews for enterprise policy controls (Intune/Entra hooks, DLP integration).
2. Flight Hub / Windows Insider blog posts and KB entries that map specific CFR rollouts and the exact behavior changes across device classes. Confirm any build/KF differences on-device via Settings → Windows Update or winver.
3. Third‑party EDR and DLP vendor updates: these vendors will need to adapt to agentic automation patterns (agent accounts, agent workspace telemetry) for enterprise detection and prevention.
4. Microsoft’s answers on rollback semantics, agent vetting and attestation processes — the practical acceptability of agentic automation in the enterprise depends on these details.

---

Conclusion​

Build 26220.7262 (KB5070303) is a compact, preview‑quality flight with a meaningful strategic signal: Microsoft is intentionally surfacing the system controls that will let Windows host agentic AI, while simultaneously improving accessibility and smoothing daily workflows. The new Experimental agentic features toggle is the most consequential user‑facing change — it turns a long‑rumored architectural direction (agents that act on your behalf) into a controllable, opt‑in test that Insiders and IT teams can evaluate.
This is an important moment: the promise of multi‑step automation delivered safely on the desktop could materially boost productivity. At the same time, it raises hard operational and security questions that must be answered before broad enterprise adoption. For Insiders and enthusiasts, this build is worth exploring on test hardware; for enterprises, the sensible posture is cautious piloting coupled with demands for stronger management, auditing and rollback guarantees before enabling agentic features at scale.

---

---

Source: Neowin Windows 11 25H2 gets big update with experimental agentic settings, more in build 26220.7262