# Microsoft closes the Windows kernel to antivirus vendors and that is a good thing

> Microsoft opens its user-mode security platform so that antivirus products stop living in the kernel. Defender, ESET and Falcon have 2027 dates.
> By José Maldonado · Cybersecurity · 2026-11-22

A faulty CrowdStrike update left some 8.5 million Windows machines in a boot loop on July 19, 2024, and paralyzed airports, hospitals and businesses. Two years and four months later, Microsoft has begun shipping the piece that prevents a repeat: a platform that lets security products do their job without putting their own code into the operating system kernel. The [Windows Latest analysis of antivirus and the kernel](https://www.windowslatest.com/2026/11/22/after-a-bad-security-update-blue-screened-millions-of-pcs-microsoft-now-lets-antivirus-run-outside-the-windows-kernel-and-windows-can-recover-if-one-breaks-startup/) details which functions the first release covers and which vendors have already committed to adopting it.

The idea is called Windows Endpoint Security Platform and it works the opposite way to the traditional model. Instead of each vendor installing its own driver and monitoring the system from the inside, Windows exposes a set of user-mode interfaces through which it delivers alerts, receives rules and enforces blocks. The security product still sees what is happening, but it no longer shares space with the code that runs memory and disk.

The initial scope is deliberately narrow. The interfaces cover the file system, processes, the registry and the object manager, which are the areas where malware detection on a desktop machine is concentrated. Anything outside that list will still need a kernel-mode component for a while, and that is the part that should not be overstated.

## What does it cover and who has signed up?

The documentation Microsoft has shared with vendors describes three capabilities of the same mechanism: filtering, notification and blocking. With them, an antivirus can observe events and decide what to do according to its own rules, without writing its own privileged code.

Vendor

Product

Announced availability

Bitdefender

Antivirus Free

Fourth quarter of 2026

Sophos

Sophos Endpoint

Fourth quarter of 2026

ESET

Endpoint Protection

February 2027

CrowdStrike

Falcon

First quarter of 2027

Microsoft

Defender

Preview in the first quarter of 2027

SentinelOne, Trellix, TrendAI

Their suites

First half of 2027

The timetable sets expectations. None of this is running generally across the enterprise estate today, and the vendor that caused the 2024 incident will be among the last to arrive. The quote that best sums up the approach is from Pavan Davuluri, head of Windows: "historically, some security protections ran in the Windows kernel, where a single failure could affect the entire device."

## Why was the kernel the problem?

A kernel-mode driver shares the most privileged space in the system. That position gives it access to everything, and it also gives it the ability to bring everything down. When that code has a bug, it is not an application that crashes: the machine crashes, and with it the ability to boot and fix the problem through the normal channels.

The user-mode model changes that equation through isolation. If a component fails, the system can stop the process and stay alive, just as happens when a browser tab hangs. The comparison is not exact, because an antivirus needs privileges a browser does not have, but it explains why a well-designed API can replace part of the work now done from the inside.

That replacement will be partial, and that is not a flaw in the announcement but a feature of the problem. CrowdStrike itself has stated that the platform "complements" the kernel protection Falcon already uses, without promising to abandon it. Anyone who reads the headline as the end of security drivers will be disappointed in 2027.

## What an API does not solve

There are three things this change does not fix, and they should be kept separate from the news. The first is that products will still need deep capabilities to detect threats that live below the file system, such as malware installed in firmware or malware that tampers with boot.

The second is that isolation reduces the damage, not the risk. A misconfigured antivirus will still block legitimate processes and a faulty detection engine will still make wrong decisions, only now with less chance of leaving the machine unable to boot.

The third is the safety net Microsoft presented in parallel. Quick Machine Recovery and Cloud Rebuild allow a machine that will not boot to be recovered without reinstalling from scratch and without losing files, and that pair of features is the real answer to a failure like the one in 2024. Having recovery tools is not the same as breaking nothing, and the official message blends the two ideas too easily.

## The timetable no one controls

The most optimistic date in the announcement puts the first antivirus running outside the kernel at the end of 2026. The most conservative puts it in the first half of 2027, and in between there is a question no press release answers: what happens with the old versions of those products, which are the ones installed on most machines.

That question matters because the enterprise estate does not update at the pace of announcements. The [management of November patches in Windows](https://observatorioblockchain.com/en/cybersecurity/patch-tuesday-noviembre-2026-outlook-cero-dia/) already showed this month how real adoption works, with updates installed in dribs and drabs and machines left on the previous version for months. Something similar will happen with antivirus products, and the driver will remain in the kernel of many machines while the new platform is rolled out in phases.

For an IT manager, the practical consequence is that the decision to adopt the platform cannot be made by looking only at the vendor's catalog. You have to check which specific version supports it, on which operating system and with what requirements, because the initial coverage is limited to very specific functions and the rest still depends on privileged components.

## The lesson of July 2024

The announcement is better understood as an admission than as an innovation. For twenty years, the Windows security ecosystem grew inward, with each vendor adding privileged layers to see more than its competitor. The result was an operating system whose stability depended on the quality of third-party code, and July 2024 put a price on that dependence.

The direction Microsoft has taken is the right one and also the slowest, because it requires each vendor to rewrite part of its product and each customer to validate the change in its own environment. The figures from that incident remain the reference: less than 1% of the world's Windows estate, and still enough to halt activity across entire sectors for days.
