← Blog

How to roll back a Windows Server update

Windows updates fix real security holes, so skipping them isn’t the answer. But now and then, one breaks something. On Windows Server it happened two months running in 2026: September’s update made Remote Desktop unstable on some servers, and October’s caused trouble for on-premises Exchange.

When it happens, you have more options than “uninstall the update”. This post goes through them in order, from the least drastic to the last resort. Each step is general advice that applies to any update problem. Where one of the two 2026 cases shows the step in practice, it’s set apart in a Real example box.

1. Confirm it’s the update

Start by checking that the problem began after an update was installed, and which update that was. This lists the most recently installed updates:

Get-HotFix | Sort-Object InstalledOn -Descending | Select-Object -First 5 HotFixID, Description, InstalledOn

This shows the exact build the server is on:

Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion' | Select-Object CurrentBuild, UBR

Put the two numbers together (for example, 20348 and 5761 make build 20348.5761) and match it to the “OS Build” in the title of Microsoft’s update articles for your version.

Then check whether Microsoft has acknowledged the problem. There are two places to look: the Windows release health pages, which list known issues for each Windows Server version, and the “Known issues” section of the update’s own KB article. Microsoft gives each known issue a status. The three you’ll see most are:

  • Confirmed: Microsoft has verified the problem and is working on it.
  • Mitigated: a workaround is available, but there’s no fix yet.
  • Resolved: a later update fixes it.

If your problem isn’t listed, it may still be the update, but check the event logs and anything else that changed around the same time before you start undoing things.

Real example, October 2026: on 16 October, the Exchange Team published a registry workaround, and Microsoft’s release health pages then listed the problem as “Mitigated”: a workaround, but no fix. The fix arrived on 10 November, four weeks after the October Windows update, in KB5131080 for Server 2025, KB5131090 for Server 2022 and the matching updates for 2019 and 2016.

2. Apply Microsoft’s workaround

If Microsoft offers a workaround, it’s usually a registry or configuration change. Read what it gives up before applying it. Updates often change a default for security reasons, and the workaround can partly undo that. Note what you changed, so you can reverse it once a proper fix arrives.

Real example, October 2026: the October Windows update set a default limit of 512 simultaneous RPC Endpoint Mapper lookups, and busy Exchange servers could hit it. The workaround was a registry value, EpmapSavedContextCap, that raised the limit and let Endpoint Mapper use more memory. The November Windows Server updates fixed the problem, and Microsoft says the value can stay or be removed. We covered it in October’s Windows Server updates.

3. Use Known Issue Rollback

A Known Issue Rollback (KIR) switches off only the change that caused the problem. Everything else in the update, including its security fixes, stays in place. A KIR can only switch off a non-security change, so not every problem gets one. Microsoft announces a KIR in the release health entry or the KB article for the update, with a separate .msi package for each Windows version. Sometimes the announcement appears only in the release health section of the Microsoft 365 admin center, which needs certain enterprise licences.

On servers, don’t count on it applying by itself. Apply it with Group Policy:

  1. Download the KIR package for your exact Windows Server version and install it on the machine you manage Group Policy from. This adds a policy template. If your domain uses a central store, copy the new .admx and .adml files there.
  2. Create or edit a Group Policy Object linked to the affected servers. Under Computer Configuration > Administrative Templates, open the folder named after the rollback (it starts with the update’s KB number) and set the policy inside it to Disabled. It sounds backwards, but Disabled is what turns the rollback on. The announcement gives the exact path.
  3. Run gpupdate /force on the affected servers, then restart them.

On a standalone server, set the same policy in the Local Group Policy Editor (gpedit.msc) after installing the package, then restart. Server Core doesn’t include the Local Group Policy Editor, so use a domain Group Policy Object there. If it isn’t in a domain, go to the next step. Once a later update fixes the problem, remove the policy.

Real example, September 2026: after the September update (KB5122871 on Server 2025, KB5122882 on Server 2022, KB5122876 on Server 2019 and KB5123099 on Server 2016), Remote Desktop connections could drop after a few minutes, sign-ins could fail, and some servers hung while Remote Desktop was being configured. Within days, Microsoft published Known Issue Rollback packages for each version, such as Windows Server 2025 KB5122871 260911_18472 Known Issue Rollback.msi. Applying one restored Remote Desktop without removing September’s security fixes.

4. Install the out-of-band fix

Sometimes Microsoft releases a fix outside the normal monthly cycle, known as an out-of-band update. It’s often available only from the Microsoft Update Catalog, not through Windows Update or WSUS. Download the .msu file for your version, install it from an elevated prompt using the command in its KB article (or by double-clicking it on a server with a desktop), restart, and check that the build number has gone up. On Server 2025, the Catalog may offer more than one file for an update; follow the installation steps in its KB article.

Most out-of-band updates are cumulative, so they include that month’s security fixes as well as the repair. The KB article says whether yours is. If you applied a Known Issue Rollback, remove its policy afterwards.

Real example, September 2026: on 14 September, six days after the monthly update, Microsoft released out-of-band fixes for Remote Desktop: KB5129235 for Server 2025, KB5129237 for Server 2022, KB5129238 for Server 2019 and KB5129239 for Server 2016. For Server 2019, 2022 and 2025 they weren’t offered through Windows Update; you had to download them from the Microsoft Update Catalog.

5. Last resort: uninstall the update

Uninstalling the update removes everything it brought, including that month’s security fixes, and leaves the server exposed until you reinstall it. Use it only when the steps above don’t work or aren’t available yet, and plan to put the update back as soon as there’s a fix.

On Server 2019 and later, the monthly update comes bundled with an update to the servicing stack, the part of Windows that installs updates. Microsoft says the usual wusa /uninstall command won’t work on that bundle. Server 2016 still gets its servicing stack update separately, so wusa /uninstall /kb:<number> works there, but the DISM method below works on every version. Run these commands in an elevated PowerShell or Command Prompt window. First, find the update’s package name:

DISM /Online /Get-Packages /Format:Table | findstr RollupFix

Find the Package_for_RollupFix entry that matches the build you found in step 1, and check that its state is Installed. Don’t remove an older or unrelated package. Then remove it and restart:

DISM /Online /Remove-Package /PackageName:<package name>

The servicing stack part stays installed; Microsoft doesn’t allow it to be removed, and that’s normal. Some updates can’t be removed, for example once Windows has cleaned up the files needed to uninstall them. If DISM refuses, don’t force it.

Then stop the update from reinstalling itself at the next check. On Server Core, open sconfig, choose option 5 (Update setting) and switch to manual or download only. Elsewhere, use the Group Policy setting “Configure Automatic Updates”, or hold the update back in WSUS or whatever tool you use to manage updates. Switch automatic updates back on once the fix is out.

Real example, September 2026: some administrators uninstalled the September update to get Remote Desktop back. It worked, but it left those servers without September’s security fixes until they installed the out-of-band update. The Known Issue Rollback and the out-of-band fix solved the same problem without that gap.

6. If the server won’t boot

If an update leaves a server unable to start, boot it from Windows Server installation media for the same version. On a virtual machine, attach the ISO through its console and make sure the machine boots from it first. Choose Repair your computer (Repair my PC on Server 2025 media), then Troubleshoot and Command Prompt. If the system drive uses BitLocker, you’ll need its recovery key. Drive letters often change in recovery, so find the Windows drive first: run bcdedit and look for the drive letter next to osdevice, or check for a Windows folder with dir D:\.

Then undo the update’s unfinished changes, using the drive letter you found:

DISM /Image:D:\ /Cleanup-Image /RevertPendingActions

Microsoft says to use this only on a system that failed to boot. Restart and try again. If the server still won’t start, go back to the command prompt and list the installed packages. Find the Package_for_RollupFix entry that matches the update installed just before the failure, and check its version and state before removing it. If you can’t identify the right package, stop and restore from a backup or get help.

DISM /Image:D:\ /Get-Packages /Format:Table
DISM /Image:D:\ /Remove-Package /PackageName:<package name>

These commands are general guidance, provided as-is. If you’re not sure, ask whoever manages your server before running them.

Before the next update

  • Take a backup or snapshot before updating anything important, and make sure you can restore it. Microsoft doesn’t support virtual machine snapshots of Exchange servers, so use a proper backup for those.
  • Make sure you have console access that doesn’t depend on Remote Desktop, in case Remote Desktop is what breaks.
  • Update a less critical server first when you can, and give it a few days before the rest.
  • Check the release health pages a few days after each monthly update for newly acknowledged problems.
  • Note the build number before and after updating, so you know exactly what changed.

What we do

We hold each month’s updates for a short watch period while we check for reported problems, then roll them out in each client’s agreed maintenance window. Critical security fixes don’t wait. If an update does cause trouble, we work through the steps above, so our clients don’t have to.

Sources

This post is general information, not professional advice for your specific setup.

Rather not do this yourself?

Dealing with updates that go wrong, so you don't have to, is part of what our Server Management service takes off your hands.

See how it works →