The path changes between the check and the action

Cyderes reproduced RoguePlanet on a fully patched Windows 11 host, starting from a standard user account, and Defender’s own scanner carried the attacker to SYSTEM 1. Nothing in the chain was a memory corruption bug. The engine resolved a file path, decided the file there was malicious, then acted on that path, and an attacker who controlled the directory had a say in what sat there the second time it was read.

The part I keep returning to is where the artifact landed. Defender wrote its quarantine copy into a directory the attacker controlled, the attacker overwrote that copy with a payload, and the Windows Error Reporting task that runs as SYSTEM picked the payload up and executed it 1. Directory junctions, opportunistic locks, a volume shadow copy and a scheduled task, each one a documented Windows feature, arranged so the privileged scanner did the moving.

Microsoft files the weakness as improper link resolution before file access, CWE-59 2. Cyderes recorded that a Defender signature covers the compiled sample and that minor source changes defeat it 1. A detection name does not fix an ordering problem.

The record carries two scores for one race

Microsoft lists the flaw as Important, with a base score of 7.8 and the exploitability index at Exploitation More Likely. The exploitation field on that same record reads No 2. NVD scores the entry at 7.0 3. The vector explains the split, with Microsoft rating attack complexity low and NVD rating it high 3. Both describe the same race and disagree on how hard it is to win. Cyderes quotes the researcher calling the current exploit hit or miss, with 100 percent success on some test machines and failure on others 1.

The two fields answer different questions. The exploitability index predicts whether someone will build an exploit; the exploitation field records what has been seen. Neither says anything about the build a given host runs.

The KEV catalog released on 13 November 2026 carries 1,781 entries and does not include this CVE 4. NVD marks the record as modified after enrichment, with a last-modified date in October 3, so the entry keeps moving.

The fix arrived as an engine build

Microsoft’s guide names the version boundaries: 1.1.26050.11 as the last engine release affected, and 1.1.26060.3008 as the first with the flaw addressed 2. Neither number is a Windows cumulative update. Engine updates ship inside the security intelligence package, which Microsoft’s documentation lists as a monthly release, and the vulnerability guide adds “or as needed” 52. Platform updates are a separate monthly package 5.

That split is the operational point. A host can take every Windows update on the calendar and still run an engine from before the fix, because the engine moves on its own schedule through its own package.

The status output exposes both numbers side by side. Get-MpComputerStatus returns AMEngineVersion and AMProductVersion as separate fields 6, so I can read the engine without inferring it from a platform build. When I checked Microsoft’s update page, the published engine was 1.1.26100.4 with platform 4.18.26100.4 7. Engine 1.1.26100.4 is above 1.1.26060.3008, the first build with the flaw addressed 2; the platform number belongs to a different package, so I read it separately. Malwarebytes reported that build shipping in July as the fix 8, and that is the number I compare against.

What I check before I call a host patched

The Defender stack is mine to track, engine and version follow-up included, and the deployment itself runs on automation. That puts the version number in my hands, where I have to read it off a host without asking anyone else.

First, the engine build has to be at or above the fixed release. Second, the distribution path has to be working. Microsoft answers “why is no action required to install this update” by pointing at automatic updating, then recommends that customers “regularly verify whether software distribution, such as the automatic deployment of Microsoft Malware Protection Engine updates and malware definitions, is working as expected in their environment” 2. That sentence makes the automatic path the control and the checking of it my job.

Third, the engine has to be the one in use. The same guide notes that vulnerability scanners find Defender binaries and version numbers on disk even when Defender is disabled, and that a system with Defender disabled is not in an exploitable state 2. A version scan can flag a host that cannot be exploited and miss a host that can, depending on which file it reads.

One mechanism leaves hosts behind. Endpoint DLP or Device Control monitoring running processes can hold a platform update back, and Defender retries the update after a reboot or once those services stop 5. A host ends up on an older build without anyone deciding to leave it there.

The version line tells me which bug is closed

In August, a researcher published ShieldBreak and described it as a full bypass of the July fix, tested on Windows 11 25H2 including Canary builds and on Windows Server 2025, and named Windows 10 as vulnerable but outside the PoC’s support 9. Tanium’s writeup describes the bypass as reopening the same race through a different technique, a rogue cloud storage provider attached to a crafted placeholder file. The same page covers the 11 August 2026 disclosure and records that, at the time it was written, no Microsoft update had closed the bypass 10. The Microsoft guide for the CVE still names the July engine build as the version with the flaw addressed 2.

All of that fits inside what an engine version can tell me. It tells me which published bug I have closed. It says nothing about whether the pattern behind the bug has closed. That pattern is narrow. A privileged process resolves a path it does not own, then acts on that path later.

What I take to the next agent

The endpoint agents, backup tools, log collectors and installers I check all run with privilege and read paths a normal user can change. Several of them resolve a path more than once per operation. I ask three questions of each one now: what does it resolve, when does it act, and who executes the result. The third question is the one that turned a file scan into a SYSTEM shell, because the answer was a scheduled task nobody was watching.

I record the engine build and the platform build on the same line in my notes, and the pairing is what tells me whether the engine in front of me is the one that closes this bug.


  1. Jayapaul, R., & Ramesh, R. (2026, June 10). RoguePlanet: Windows zero-day that weaponizes Defender’s quarantine pipeline. Cyderes. https://www.cyderes.com/howler-cell/rogueplanet-windows-zero-day ↩︎ ↩︎ ↩︎ ↩︎

  2. Microsoft Corporation. (2026, July 8). CVE-2026-50656: Microsoft Defender elevation of privilege vulnerability. Microsoft Security Update Guide. https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-50656 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎

  3. National Vulnerability Database. (2026, October 25). CVE-2026-50656 detail. National Institute of Standards and Technology. https://nvd.nist.gov/vuln/detail/CVE-2026-50656 ↩︎ ↩︎ ↩︎

  4. Cybersecurity and Infrastructure Security Agency. (2026, November 13). Known exploited vulnerabilities catalog (catalog version 2026.11.13). https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json ↩︎

  5. Microsoft Corporation. (2026, May 14). Microsoft Defender Antivirus security intelligence and product updates and support. Microsoft Learn. https://learn.microsoft.com/en-us/defender-endpoint/microsoft-defender-antivirus-updates ↩︎ ↩︎ ↩︎

  6. Microsoft Corporation. (n.d.). Get-MpComputerStatus. Microsoft Learn. https://learn.microsoft.com/en-us/powershell/module/defender/get-mpcomputerstatus ↩︎

  7. Microsoft Corporation. (2026, November 16). Latest security intelligence updates for Microsoft Defender Antivirus and other Microsoft antimalware. Microsoft Security Intelligence. https://www.microsoft.com/en-us/wdsi/defenderupdates ↩︎

  8. Arntz, P. (2026, July 9). Microsoft fixes RoguePlanet zero-day in Defender. Malwarebytes. https://www.malwarebytes.com/blog/news/2026/07/microsoft-fixes-rogueplanet-zero-day-in-defender ↩︎

  9. Paganini, P. (2026, August 12). ShieldBreak: New Windows zero-day bypasses Microsoft’s RoguePlanet patch. Security Affairs. https://securityaffairs.com/197063/hacking/shieldbreak-new-windows-zero-day-bypasses-microsofts-rogueplanet-patch.html ↩︎

  10. Guardian Research Team. (2026). ShieldBreak: The Windows Defender 0-day with no patch, and what to do about it. Tanium. https://www.tanium.com/blog/shieldbreak-mitigation ↩︎