Back to Blog

Administrator protection: the baseline switched it on 35 days before the announcement

Room with rows of identical desktop computers on white desks

If your organisation applies the security baselines from the Security Compliance Toolkit and nobody has touched the policy, Administrator protection has been on across your machines since the October security update, which shipped on the 13th. The Windows 11 release health page says so: the feature is "enabled by default" in the 24H2, 25H2 and 26H2 baselines, "unless your organization has configured the policy differently". Microsoft flagged it on that same 13 October through the message center — MC1506291, tagged "admin impact", naming the two UAC policies involved — and put it on the Ignite stage on 17 November: "With Windows administrator protection, now generally available, IT can enable local administrators to run with standard user privileges by default." Between the notice and the stage there are 35 days. What this post is about is what happens in a fleet during those 35 days, because the notice was published and most of the coverage arrived with the announcement.

That this is not a technicality is proved by the trail it left. On 13 October at 09:43 Pacific time Microsoft opened a known issue: non-admin users could not connect over Remote Desktop to devices with Administrator protection enabled. The error, 0xC000000D, and the message, "The token that was passed to the function is invalid." It originated in a preview update from 22 September, KB5124010, and was closed on 28 October with KB5127936. The bug is fixed. What stands is the scope Microsoft gave it: it primarily affects enterprise-managed devices used as Remote Desktop hosts with the feature enabled, "regardless of how the feature was enabled". That "how" is a link to the list of methods — Intune, CSP, Group Policy and the Windows Security app — and the baseline configures exactly those policies in one go for the whole group you assign it to.

The two facts that decide which side you are on sit one after the other in the same requirements section: the feature is "off by default" and, in the note beside it, "Administrator protection becomes enabled by default with the October 2026 Windows security update" if you use the baselines. Windows does not ship with it on; the security template you apply turns it on. If you do not apply the Security Compliance Toolkit baselines, you are still off and the decision is entirely ahead of you. If you do apply them — which is what Microsoft publishes them for — the decision is already made and the behaviour change has been on your laptops for five weeks.

The admin token comes from an account you never created

The quick reading is "UAC now asks for your face instead of the Yes button". It is not that. What changed is where the privileges come from. The documentation describes it this way: when elevation is needed, "Windows uses a hidden, system-generated, profile-separated user account to create an isolated admin token. This token is issued to the requesting process and is destroyed once the process ends." In the article where Microsoft introduced the model that account has a name — System Managed Administrator Account, SMAA — defined as "a local administrator account which is linked to a specific standard user account", and "upon elevation, if a SMAA does not exist already it is created". Two lifetimes worth not mixing up: the token dies with the process; the account and its profile stay, which is why whatever you leave in the elevated profile is still there tomorrow. The elevated process, technically, is no longer you: it is a neighbour of yours with a different profile and a different SID.

We think the change is right, and that is worth saying before listing what it breaks. With Administrator protection, auto-elevation is gone: no application elevates on its own, every one of them has to ask with a human present. Microsoft put numbers on that when it introduced the model in January 2025: the surface being removed is "roughly 92 auto-elevating COM interfaces, 11 DLL Hijacks, and 23 auto-elevating apps". For anyone who has read an incident response report, that list is half the privilege ladder of any workstation intrusion. The idea was never the problem. The problem is that an identity change dressed up as a security checkbox reaches the fleet through the same door the patches use, and nobody treats it as what it is.

The whole breakage list follows from the root cause

The useful part of understanding the mechanism is that there is nothing to memorise. If the elevated process runs under another account with another profile, anything that depends on "what is in my session" is no longer there. These are the consequences the documentation itself lists, ordered by how many people they annoy:

What you stop having Why What to do
Single sign-on inside the elevated app The standard session's credentials "aren't available to the elevated session" Expect a second domain or cloud sign-in inside that window
Network drives from the elevated process The other profile does not have your connections mounted The docs say it literally: copy installation files to a local drive before elevating
The settings of apps you run elevated "Settings data for applications don't carry over across the regular and the elevated profiles" Only elevate what you truly need; that is Microsoft's advice, not ours
WSL and elevated dev environments The distro lives in the user profile, and the elevated one is a different profile It has to be reinstalled and configured separately in the elevated profile
Silent elevations by scripts and installers "Users need to interactively authorize every admin operation" Inventory what elevates in your fleet with nobody present (that is the real work)

Two more things that were broken and no longer are, in case anyone is still carrying them: some applications did not show up in the Start menu after installing, and others using Edge extensions asked for elevated permissions even when launched normally. Both were closed with that same September KB5124010. If either rings a bell, that machine has not been kept up to date for two months, which is a different conversation.

Where the documentation itself says not to enable it

This is the most honest part of the page and the first one any summary drops. Under the heading "when not to enable this feature", Microsoft lists: Windows 365 Cloud PCs, Azure Virtual Desktop session hosts, devices that require Hyper-V, apps that access shared files across profiles and — the one that matters in a managed fleet — devices managed with LAPS, the Local Administrator Password Solution. It adds, in the limitations list, that roaming profiles and backup admins are not supported either.

If you manage real workplaces, there is an awkward conflict with no clean way out: the baseline turns on something the same documentation advises against on LAPS-managed devices, and LAPS is exactly what you deploy so you do not reuse the local admin password across two hundred laptops. It is not a Microsoft mistake or a trap: they are two defensive layers that overlap, and you have to decide which one wins in each group of machines. What does not work is finding out on the day a technician cannot get in.

The default that shows up in remote administration

There is a policy with a very long name that decides whether a domain administrator logging in remotely arrives elevated or not. It is called "Allow remote logon with elevated privileges for domain users in the local Administrators group when Administrator protection is enabled", and its default value is disabled: domain users in the local Administrators group "perform remote logons without elevated privileges". Enabling it restores classic UAC behaviour. There is also a granular variant that grants remote elevation only to designated users or groups.

Our recommendation, and we will commit to it: do not restore classic behaviour for everybody. The temptation is obvious, because the long-standing support procedure stops working on the first try and the global switch fixes it in a minute. But that switch disables half the value of the feature precisely on the accounts with the most access. The granular variant — remote elevation only for the group of technicians that needs it — costs an afternoon and keeps the rest. And if someone asks you to disable Administrator protection entirely to resolve a ticket, put it in writing that this was Microsoft's workaround for a bug that was fixed three weeks ago.

One fleet for patching, two for elevating

The same Ignite announcement carries another sentence that will end up in a lot of slide decks: "For the first time, endpoint teams can manage servers and PCs as one Windows fleet with Intune and Windows Autopatch." It is true and it is a good headline. It is worth reading next to what the Administrator protection page says: support for Windows Server editions "is planned for a future release". The fleet is one for inventory and patching, and it is still two for the privilege model. If somebody leaves that session thinking servers stop having standing administrators, correct them before it ends up in a plan.

Four checks to know today where you stand

  • Do you apply the Security Compliance Toolkit baselines? If the answer is yes and nobody has touched the policy, assume it has been on since October. It is the check that requires opening no console at all.
  • Look at the policy, not at the feeling. In Intune, settings catalog, category Local Policies Security Options, setting "User Account Control Type Of Admin Approval Mode". In Group Policy, secpol.msc → Local Policies → Security Options → "User Account Control: Configure type of Admin Approval Mode", with the value "Admin Approval Mode with Administrator protection". A reboot is required for it to take effect, both when turning it on and when turning it off.
  • Check what it asks of your users. The 26H2 baseline does not stop at switching it on: it also sets the prompt type to "prompt for credentials on the secure desktop". People clicking a button is not the same as people typing their password every time an installer asks for permission. That user experience decision was made by a template, and it is changed with the second setting in the catalog.
  • Measure before you widen. The feature ships two new events under the Microsoft-Windows-LUA provider: 15031 when an elevation is granted and 15032 when it is denied, fails or times out. You capture them with logman start AdminProtectionTrace -p {93c05d69-51a3-485e-877f-1806a8731346} -ets, and each record carries the SID of whoever elevated, the application and its path, the outcome, the managed account that was used and the authentication method. A pilot group running that for two weeks is worth more than any list of "compatible" applications.

The change is still yours even when the vendor signs it

The study we have been citing for years — Oppenheimer, Ganapathi and Patterson, USENIX 2003, on why internet services fail — put operator configuration changes at the top of the causes list, well ahead of hardware. Twenty-three years later the script has turned subtler: the change is no longer typed by someone on a Friday, it arrives inside a template you apply out of hygiene and that is usually right. It is still a configuration change in production. Microsoft signing it does not make it something else, it only means nobody writes it down in the change log.

That is why the rollout order matters more than the date. First the inventory of what elevates with nobody present, which is where the surprises live; then the technicians, who are the people who can describe a failure properly; users last. It is the same discipline we described when we went through the data on testing the live system instead of trusting paper, and the same reason we keep insisting that a console that manages the whole fleet is the most delicate piece you own: what changes from there changes everywhere at once, for better and for worse. On versions, this applies to Windows 11 in 24H2, 25H2 and 26H2 — on 24H2 servicing has already ended for Home and Pro, and not for every edition sharing its build — and on servers it does not exist yet.

Sources (verified at origin on 19/11/2026): the general availability announcement and the sentence about servers and PCs as one fleet come from Pavan Davuluri's 17/11/2026 article on the Windows blog, "Microsoft Ignite 2026: Building Windows for the next era of work". How it works (hidden system-generated profile-separated account, token destroyed when the process ends, no auto-elevations, Windows Hello authorisation), the system requirements and KB5120998, the Intune and Group Policy settings, events 15031 and 15032 with their GUID and the logman command, the "when not to enable" list (Cloud PCs, AVD, Hyper-V, apps sharing files across profiles, LAPS), the limitations on SSO, network drives, settings across profiles and WSL, the remote logon policies with their default value and the plan for Windows Server support all come from "Administrator protection" on Microsoft Learn (page dated 13/10/2026). The note about baselines, the 0xC000000D issue with its dates (opened 13/10/2026, originating in KB5124010 of 22/09/2026, resolved 28/10/2026 with KB5127936) and the "regardless of how the feature was enabled" clause are from the Windows 11, version 26H2 release health page, updated 18/11/2026, which is also where the date of the October security update (13/10/2026) comes from. That the baseline also sets the prompt to "prompt for credentials on the secure desktop", and that the 24H2, 25H2 and 26H2 baselines include settings enabling the feature, come from the Windows 11 26H2 security baseline, published 29/09/2026 and updated to a v2 package on 08/10/2026 which is the one adding the Administrator protection settings, per the update note on that same entry. The notice to administrators is MC1506291 in the Microsoft 365 message center, "Manage Administrator protection within Windows 11 security baseline", dated 13/10/2026 and tagged with admin impact. The removed-surface figures — 92 COM interfaces, 11 DLL hijacks, 23 applications — along with the name System Managed Administrator Account (SMAA), its definition as a local administrator account linked to a specific standard user account and the fact that it is created on first elevation if it does not exist, are from "Evolving the Windows User Model – introducing Administrator protection", Microsoft security blog, 28/01/2025. The outage causes study is Oppenheimer, Ganapathi and Patterson, "Why Do Internet Services Fail, and What Can Be Done About It?", USENIX 2003. Translations of the quotes are ours; the English originals are behind the links. The 35 days between 13 October and 17 November are our own arithmetic. Photograph: computer room at the University of Warwick, by Mike1024, public domain via Wikimedia Commons. We are not a reseller of Microsoft licences or anybody else's: we charge for designing and operating, not for what you subscribe to.

Do you know what elevates in your fleet today with nobody present?

At everyWAN we build the modern workplace by deciding these things before a template decides them, and we back them with 24x7 IT support that knows what to do when remote elevation stops working at three in the morning. If you want to review your elevation policies, your pilot group and your emergency access procedure, write to us.

Talk to everyWAN

Tags:

Share:

Subscribe to our newsletter

To receive IT stories, everyWAN news and exclusive subscriber offers, sign up to our mailing list

Minorisa de Sistemas Informaticos y Gestión S.L. © 2026
everyWAN
everyWAN