CISA added CVE-2026-32202 to the Known Exploited Vulnerabilities catalog on 28 April, with a due date of 12 May 1. Microsoft had shipped a fix two weeks earlier, on 14 April 2. On 28 July, Microsoft re-released the CVE with a note telling administrators to install the July 2026 updates 2. Three dates for one bug, and a fleet can meet the middle one on a build the vendor later moved past.
The advisory files the flaw under spoofing, at a CVSS 3.1 base of 4.3 and a severity of Important 2. The attack takes a credential, and that is the part I plan around. The exploitability section, which I pulled on 31 October, still reads Exploited: Yes with an assessment of Exploitation Detected 2.
What the April update left running
The flaw exists because of where Microsoft put the check. Akamai’s researchers found the coercion path while they tested the February update for CVE-2026-21510, the APT28 shortcut bug 3. That update routed the Control Panel launch through a new object, ControlPanelLinkSite, so ShellExecute could ask SmartScreen about the applet before running it 3. The check runs inside ShellExecuteExW, at the end of the chain.
The chain starts earlier. Explorer renders a folder, asks shell32 for an icon for the Control Panel item, and shell32 resolves the path to fetch it. Akamai traced the first trigger to PathFileExistsW inside GetModuleMapped, called from CControlPanelFolder::GetUIObjectOf 3. When the path is a UNC path, Windows opens an SMB connection to whatever host answers, the connection triggers an NTLM handshake, and the victim’s Net-NTLMv2 hash travels to the attacker, who relays it or cracks it offline 3. Nobody clicks, because rendering the folder does the work.
The file carries no payload. The LNK holds a header and a LinkTargetIDList: the Control Panel CLSID, an item for all control panel items, and a third item that embeds the UNC path to a .cpl 3. There is no DLL to sign and no macro to inspect.
Two fixes in one record
Revision 2.0 of the advisory, dated 28 July, says the CVE has been “re-released to comprehensively address the vulnerability identified by CVE-2026-32202” and recommends the July 2026 updates 2. That sentence is the whole revision note, and it names a later update, which changes the build I ask for.
The KEV entry holds the April timing: added 28 April, due 12 May, ransomware use unknown, forensic triage set to No 1. Those fields do not move when a vendor revises a CVE, so a report built from them reads this flaw as closed in May while MSRC’s record points at a later update. Both pages are correct about their own scope, and the gap between them lands on the analyst.
The build I ask for
NVD’s configuration list gives me the branch builds: Windows 11 24H2 below 10.0.26100.8875, Windows 10 22H2 below 10.0.19045.7548, Server 2022 below 10.0.20348.5386 4. Those numbers tell me which hosts sit below the fixed line. They do not tell me which month’s cumulative update put a host above it, and with a re-release in the record that distinction decides my answer. I treat the July cumulative as the floor on each branch and compare the fleet against that.
Two questions go in the ticket: which builds sit on each branch, and which of those builds took a July update. The second question is the one that catches a machine patched in April and left alone since.
Which hosts can send NTLM out
A patched build stops the coercion. Whether a hash then leaves the network is a policy question, and Windows ships a setting for it. Network security: Restrict NTLM: Outgoing NTLM traffic to remote servers takes three values, Allow all, Audit all and Deny all, and leaving it undefined behaves like Allow all 5. Microsoft’s best practice on that page is to run Audit all first, read the events to learn which servers receive NTLM, add those to the exception list, then switch to Deny all 5.
I gave that sequence three weeks. Audit mode ran on a pilot ring, I built the exception list from what appeared in the logs, then switched to Deny all. A deny that lands before the exception list exists breaks mapped drives on the first Monday of a term.
The other half is SMB egress. A workstation has no business opening TCP 445 to an internet address, and blocking that egress costs nothing until someone maps a drive to a host outside the exception list. Cynet’s advisory recommends restricting unnecessary outbound SMB traffic, and the word carrying the weight in that sentence is unnecessary 6. In our environment the block covers internet egress, and the file servers stay reachable.
How I hunt LNK and CPL execution now
For LNK and CPL activity I run three queries. The first pulls file events for .lnk files written into Downloads, temp directories, archive extraction folders and removable media. The second looks for control.exe started with a .cpl argument, and I read the ones where explorer.exe is the parent first. The third watches for explorer.exe opening a connection to a public address on port 445.
That third query is the signal I lean on. A laptop on a campus network has no reason to open SMB to a public IP, and the event fires whether or not the user clicked and whether or not the CPL ever loaded. By the time it fires the hash has left, so the query supports containment rather than proving a compromise.
A relayed authentication shows up on the system that accepted it, so the workstation that sent it is the wrong place to look for success. A search for NTLM network logons into servers from workstation subnets with no reason to reach them is where a successful relay becomes visible.
What I send to the systems team
The message stays short: the July build per branch, the hosts below it, the current value of the outgoing NTLM setting, and the egress question. Deployment is automated in our environment, so what I add is the order and the reasoning behind it. July builds on the branches still carrying the April version, then the NTLM audit, then the egress rule.
I also add a line about the re-release. It explains why a CVE from April sits in a ticket in October, and it keeps the ticket from being closed against a May compliance date that never described the final fix.
The July build is what goes in the ticket, and the NTLM policy is the part I can close without waiting on a deployment.
-
Cybersecurity and Infrastructure Security Agency. (2026, October 30). Known exploited vulnerabilities catalog [Data set, catalog version 2026.10.30]. https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json ↩︎ ↩︎
-
Microsoft. (2026, April 14). CVE-2026-32202: Windows Shell spoofing vulnerability [Security Update Guide, revision 2.0 dated 2026-07-28]. Microsoft Security Response Center. https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-32202 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
Dahan, M. (2026, April 23). A shortcut to coercion: Incomplete patch of APT28’s zero-day leads to CVE-2026-32202. Akamai. https://www.akamai.com/blog/security-research/incomplete-patch-apt28s-zero-day-cve-2026-32202 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
-
National Institute of Standards and Technology. (2026, April 14). CVE-2026-32202 detail. National Vulnerability Database. https://nvd.nist.gov/vuln/detail/CVE-2026-32202 ↩︎
-
Microsoft. (2022, June 14). Network security: Restrict NTLM: Outgoing NTLM traffic to remote servers. Microsoft Learn. https://learn.microsoft.com/en-us/previous-versions/windows/it-pro/windows-10/security/threat-protection/security-policy-settings/network-security-restrict-ntlm-outgoing-ntlm-traffic-to-remote-servers ↩︎ ↩︎
-
Veltzman, I., & Malyutin, M. (2026, April 30). Emerging threat advisory: Windows Shell spoofing vulnerability CVE-2026-32202. Cynet. https://www.cynet.com/blog/emerging-threat-advisory-windows-shell-spoofing-vulnerability-cve-2026-32202/ ↩︎