Blog/Vulnerabilities

Windows Vulnerability Exposure: What the Exploited Catalog Tells an Administrator

Microsoft fills more of CISA's exploited catalog than any other vendor. What those entries have in common, and the Windows settings that change the outcome.

CyberWatch AILast reviewed 21 November 2026 · 9 min read
An IT administrator seated at a desk with two monitors and a keyboard

Download CISA's Known Exploited Vulnerabilities catalog and count the entries by vendor. As it stood on 20 November 2026, with 1,790 entries in total, Microsoft accounted for 391 of them. No other vendor is close, and 117 of those entries are flagged as having known use in ransomware campaigns.

That is not, by itself, a verdict on Windows security. It reflects how much Windows there is, and it reflects visibility: Microsoft and the researchers who study its products generate an unusual amount of public exploitation evidence. But if you administer Windows, the catalog is the most useful document available about what actually gets used against you, and it has patterns worth reading rather than just a list worth patching.

What the Windows entries have in common

Sort those entries by product and three concentrations appear.

Kernel and driver flaws used for local privilege escalation. This is the largest group, and the repetition within it is striking. The Common Log File System driver produced three separate catalog entries in 2025: CVE-2025-29824, a use-after-free that CISA describes as allowing an authorized attacker to elevate privileges locally and which carries a known ransomware flag, followed by CVE-2025-32701 and CVE-2025-32706 within weeks. The Ancillary Function Driver for WinSock produced entries in February 2025, May 2025 and August 2026. Win32k has 25 entries across the catalog's history.

The lesson is not that these drivers are uniquely bad. Once researchers find one memory safety bug in a driver they go looking for siblings, and the attackers reading the same advisories do too. A driver in one month's updates is likely to appear again.

Server products reachable over the network. Exchange Server has 17 entries, SharePoint 15 across its two product names, Active Directory three. These produce the emergency change windows, and they share a characteristic: the product is deployed on-premises by organizations who assumed it was internal, and some portion of those deployments is published to the internet. CVE-2025-59287, a deserialization flaw in Windows Server Update Service, is a clean illustration. Microsoft's record describes it as allowing an unauthorized attacker to execute code over a network and scores it 9.8. An update service is not something most administrators think of as attack surface.

Client-side code that opens files. Office has 30 entries, Word four, Internet Explorer 36. That last count is a reminder that a component keeps producing exploited vulnerabilities long after the product it belonged to stops being a strategic concern, because the rendering engine survives inside other things.

The old-CVE problem

One pattern deserves its own note because it contradicts a comfortable assumption. Between August 2025 and May 2026, CISA added Microsoft entries with identifiers from 2007 through 2013, including CVE-2008-4250 and CVE-2008-0015. These are not newly discovered flaws. Fresh evidence of exploitation surfaced, which means somebody is still running the software and attackers know it.

If your estate contains systems that old, their age confers no protection. It makes them a known-good target.

Reading Patch Tuesday without drowning

A monthly Microsoft release is too large to assess item by item, so the filters you apply first decide whether the exercise is useful.

Start with exploitation, not severity. Microsoft publishes an Exploitability Index assessment alongside its updates with four values: Exploitation Detected, Exploitation More Likely, Exploitation Less Likely and Exploitation Unlikely. The first means Microsoft is aware of an instance of the flaw being exploited, and those items take top priority. Two caveats from Microsoft's documentation matter. Where a flaw affects several product series, the rating reflects the highest risk level across them. And while Microsoft will revise an assessment where a change is warranted, it does not update the index when later exploit code merely matches the existing assessment, or once an update would no longer be useful to customers. Treat it as a snapshot taken on release day.

Cross-check against CISA's catalog, which has a stricter bar: an assigned CVE identifier, reliable evidence of active exploitation in the wild, and a clear remediation action. Scanning and a public proof-of-concept do not qualify. Anything in both lists is not a judgment call.

Then filter by exposure, which is the information Microsoft cannot supply. A critical remote code execution flaw in a role you have not installed is not your problem. A medium-severity escalation flaw on four thousand laptops is.

Treat the severity scores as a tiebreaker, and read both of them. CVE-2020-1472, Zerologon, is the standing example of why. Microsoft's own CVSS version 3.1 assessment scores it 5.5, a Medium, because Microsoft scored it as requiring local access and low privileges. The National Vulnerability Database's assessment of the same flaw is 10.0, Critical, scored as network-accessible with no privileges required. The CVE record describes an unauthenticated attacker able to connect to a domain controller using the Netlogon Remote Protocol and obtain domain administrator access. One of those numbers would have put it in your monthly queue and the other would have put it in an emergency change. For the general version of this problem, see our vulnerability management guide.

Watch for flaws that need more than a patch. Microsoft shipped the Zerologon fix as a phased two-part rollout, with enforcement arriving separately from the initial update. Several Windows advisories work this way: the update makes a safer behavior available and a later step, or a registry setting, makes it mandatory. An organization that deployed the first part and never did the second was never protected. CISA's catalog has started recording a forensic triage requirement on some entries, which is a second task of the same kind: establish whether the asset was already compromised rather than only closing the hole.

The settings that change the outcome

Patching is a treadmill. The configuration below changes what an exploit yields, and most of it is free and already present.

Protect credentials in memory

Almost every Windows intrusion of consequence passes through credential theft from the Local Security Authority process. Two features address it and they are complementary rather than alternatives.

LSA protection runs LSASS as a protected process, which Microsoft describes as preventing nonprotected processes from reading its memory and injecting code into it. It has existed since Windows 8.1. The important operational detail is that on client devices running Windows 11 version 22H2 and later, Microsoft enables it by default when the device is a new installation rather than an upgrade, is enterprise joined, and is capable of hypervisor-protected code integrity. Audit mode is on by default from 22H2 regardless. So the task for most administrators is verification, not deployment: check for the WinInit event recording that LSASS started as a protected process with level 4, and investigate the machines where it is absent. Enabling it with the UEFI lock option stores the setting in firmware so that clearing the registry key has no effect, which is worth doing on anything you consider sensitive.

Credential Guard goes further, isolating NTLM password hashes, Kerberos ticket-granting tickets and application-stored domain credentials in a separate process protected by virtualization-based security. Microsoft states that virtualization-based security and Credential Guard are enabled by default from Windows 11 version 22H2 on devices meeting the system requirements, and that Credential Guard is supported on 64-bit Secure Boot devices only.

Both features break LSA plug-ins that are not signed appropriately, which is why audit mode exists. Use it to find the smart card drivers, cryptographic plug-ins and password filters that will fail to load before enforcing anything.

Stop sharing local administrator passwords

Microsoft's own description of the problem beats most security advice on the subject: "Local administrator accounts often share the same password across many devices, which bad actors can exploit to move laterally across your environment."

Windows LAPS rotates each device's local administrator password, stores it in Active Directory or Microsoft Entra ID, and lets authorized administrators retrieve it. Microsoft lists protection against pass-the-hash and lateral-traversal attacks as the first benefit, and the feature is available at no cost on all supported Windows platforms, built into Windows 11 23H2 and later, Windows Server 2025 and later, and earlier supported versions that received the April 2023 update.

One detail catches people out. The legacy Microsoft LAPS product is deprecated as of Windows 11 23H2 and newer versions block its installer, so an old deployment needs a migration plan rather than maintenance. Emulation mode, which honors the legacy Group Policy settings during a transition, stores passwords in clear text only and is a temporary step.

Reduce what untrusted code can do

Local privilege escalation requires untrusted code to be running somewhere, so the controls that matter most govern what runs at all: removing standing local administrator rights, blocking macros in documents from the internet, and application control where you can sustain it. Our piece on least privilege in practice covers the rights question, and our endpoint security guide the detection layer that catches what gets through.

One timing note belongs in any Windows plan: Microsoft's documentation records that Windows 10 reached the end of support on 14 October 2025, and devices still on it receive updates only through the Extended Security Updates programme. An unsupported build is not a problem patching can solve.

A monthly rhythm that survives a busy month

  1. On release day, pull the items marked Exploitation Detected and cross-reference CISA's catalog. That short list is your emergency track, and it needs a standing approval path so nobody negotiates a change window for it.
  2. Check applicability before priority. Which roles, features and versions do you actually run? Recording that you checked and the answer was no is a legitimate and auditable outcome.
  3. Split the rest by exposure, with different target timescales for internet-facing servers, internal servers and endpoints. NIST's guide to enterprise patch management planning notes that most of the difficulty here is organizational: maintenance windows, change approval and systems nobody will authorize downtime for are what block remediation. Decide those policies when you are not under pressure.
  4. Verify rather than trust the deployment report. Rescan the asset. Patches fail silently, reboots get deferred, and a machine reporting success while still running the old kernel is the normal failure mode.
  5. Close the second task. Phased enforcement steps, session invalidation after a credential-leaking flaw, and forensic checks where exploitation preceded the patch are separate pieces of work. If they live in the same ticket as the patch, they get closed with it.

Patch notifications are a phishing theme. Fake "critical Windows update" messages spike after major vulnerability news. Paste anything suspicious into CyberWatch AI Scan for a free check, and take updates only through your normal management channel.

The short version

Microsoft is the largest single source of entries in the catalog of exploited vulnerabilities, and they cluster in three places: kernel and driver flaws used to escalate privileges on endpoints, server products that turned out to be reachable from the internet, and client-side components that open files. Old identifiers keep appearing, which tells you unsupported systems are being found and used.

Prioritize by exploitation evidence and exposure rather than severity score, read both the vendor's score and the National Vulnerability Database's because they sometimes disagree sharply, and check whether an advisory needs a second action beyond the update. Then spend what is left on the configuration that decides what an exploit is worth: credential protection in memory, unique rotated local administrator passwords, and as little standing privilege as the business can live with. For flaws with no fix yet, see our guide to zero-day vulnerabilities.

Frequently asked questions

Why does Microsoft dominate CISA's exploited vulnerability catalog?

Partly scale, because Windows is everywhere, and partly visibility, because Microsoft's own products and security research produce a great deal of exploitation evidence. As of 20 November 2026 Microsoft accounted for 391 of 1,790 catalog entries, more than any other vendor.

What is the MSRC Exploitability Index?

Microsoft's four-level assessment of how likely a vulnerability is to be exploited, published alongside each update. The levels are Exploitation Detected, Exploitation More Likely, Exploitation Less Likely and Exploitation Unlikely. Where a flaw affects several products, the rating reflects the highest risk across them.

Should I prioritize by CVSS score on Patch Tuesday?

Not alone. Microsoft's own score and the National Vulnerability Database's score sometimes differ substantially for the same flaw, and neither knows what you run or what faces the internet. Exploitation status and exposure are better first filters.

Why are so many exploited Windows flaws local privilege escalation?

Because attackers reach endpoints routinely through phishing and malicious documents, and a local escalation flaw converts that foothold into control of the machine. Kernel drivers such as the Common Log File System driver appear repeatedly for this reason.

Is LSA protection something I need to turn on?

On new installations of Windows 11 version 22H2 and later that are enterprise joined and capable of hypervisor-protected code integrity, Microsoft enables it by default. The practical task is verifying it is active rather than deploying it, which you can check with the WinInit event recording that LSASS started as a protected process.

What does Windows LAPS actually fix?

Shared local administrator passwords. Microsoft's own documentation notes that local administrator accounts often share the same password across many devices, which attackers exploit to move laterally. LAPS rotates each device's password and stores it so authorized administrators can retrieve it.

How fast should Windows updates be deployed?

Set different targets by exposure and exploitation evidence rather than one number. A flaw in CISA's catalog affecting an internet-facing server needs action in days. The same month's local escalation fixes for laptops can follow a normal ring-based rollout, provided that rollout actually completes.

Why does CISA keep adding very old Microsoft flaws to the catalog?

Because evidence of exploitation continues to surface for software that is still running somewhere. Between August 2025 and May 2026 the catalog gained Microsoft entries with CVE identifiers from 2007 through 2013. If you have systems that old, their age is not protection.

What should I do about a Windows flaw marked as exploited before a patch existed?

Patch, then investigate. The update closes the entry point and does not evict an attacker who already used it. Several recent catalog entries carry an explicit forensic triage requirement for exactly this reason.

Sources

  1. Known Exploited Vulnerabilities Catalog, US Cybersecurity and Infrastructure Security Agency
  2. Microsoft Exploitability Index, Microsoft
  3. Security Update Guide, Microsoft
  4. Configure added LSA protection, Microsoft
  5. Windows LAPS Overview, Microsoft
  6. BOD 26-04: Prioritizing Security Updates Based on Risk, US Cybersecurity and Infrastructure Security Agency
  7. SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning, US National Institute of Standards and Technology
  8. CVE-2020-1472, CVE Program
  9. CVE-2025-59287, CVE Program
Share this article
LinkedIn WhatsApp X
For organizations

Your people get these messages at work too.

CyberWatch AI sends your staff realistic practice attacks, trains the gaps it finds, and shows management exactly where the organization stands.