VMware vCenter zero-day exploited for 2 months
A VMware vCenter zero-day let attackers become vCenter admin without a password. How to check your build, patch it and look for signs you were hit.

On this page 4 sections ▾
WHAT TO DO
- Patching is the only fix: Broadcom lists no workaround, the update can't go through Quick Patch, and vCenter 7.0 has no public patch.
- A fixed build doesn't mean you're clean. Check for single sign-on admin accounts nobody created and for Linux appliances or remote-access boxes that could reach vCenter unwatched.
- Affected
- vCenter 8.0, 9.0 and 9.1, including the builds inside VCF, vSphere Foundation and Telco Cloud; 7.0 and older too
- Fixed in
- vCenter 8.0 U3m, 9.0.2.0200 and 9.1.1.0200; VCF 5.2.5
- Disclosed
- Nov 19, 2026
- Exploited
- Yes, as a zero-day since early September 2026, per Google's Mandiant; not in CISA KEV as of November 19
- Workaround
- None from Broadcom
Broadcom shipped its emergency vCenter patch on November 19, and Google says a suspected China-linked espionage group had been using the vCenter zero-day behind it since early September. CVE-2026-123890 lets anyone who can reach vCenter over the network sign in as an administrator without a password, and vCenter is the console that controls every host and virtual machine in a VMware environment.
Quick facts
Mandiant says the group already knew about this flaw before Broadcom disclosed the previous vCenter login bypass, CVE-2026-59309, on July 29.
In the intrusion Google investigated, the attackers used it on a vCenter server that was already patched for that July flaw.
CrowdStrike calls the same group Warp Panda, and links it to another VMware zero-day this year: an ESX logging flaw that CISA added to its Known Exploited Vulnerabilities list on October 28.
#How the vCenter zero-day works
vCenter's single sign-on runs on the VMware Directory Service, and port 2012 is the remote procedure call interface that controls it. Normally a password login there runs a Secure Remote Password (SRP) exchange, which ends with both sides holding a shared session key that proves the login worked. CVE-2026-123890 hands that session key to someone who never logged in, so the attacker skips the password check entirely.
ELI5: explain it simply▾

THE WORDS
- Single sign-on (SSO)
- One login that unlocks many systems, so you don't sign in to each one separately.
- Session key
- A secret both sides agree on after a successful login, used to prove you're already signed in.
- RPC
- Remote procedure call: one program asking another program over the network to run something.
THE IDEA
Picture a club where the bouncer checks your ID and then hands you a wristband. This flaw is a side door where the club hands out wristbands to anyone who asks, without checking ID first. Once you're wearing one, every door inside opens.
#From a remote-support box to Active Directory
vCenter wasn't the way in. Here's how the intrusion Mandiant investigated unfolded, from Google's report as summarized by Cyber Security News:
- A remote-support applianceThe group broke in through CVE-2026-102764, a separate zero-day in BeyondTrust Remote Support, in late August.
- A Linux box nobody watchedFrom there it moved to an internal Linux appliance with no endpoint detection and response (EDR) software, got root and started looking for VMware systems.
- vCenter admin without a passwordA custom tool called BONESAW used the vCenter flaw to get in, then copied the directory database and created a new single sign-on administrator account.
- Forged loginsA second tool, SAWSNAKE, used stolen signing keys to forge SAML tokens, so the attackers could pose as an administrator without signing in or passing MFA.
- The data on the virtual disksThey took vCenter's encryption material and database credentials, reached an ESXi host and opened files inside a Windows virtual disk: Active Directory data, registry files and DNS data. Then they copied an archive out.
That last step is why vCenter is such a prize. Whoever controls the hypervisor can read a virtual machine's disk from the outside, so the Active Directory data came off a Windows server's disk without anyone logging in to Windows or getting past the security software running inside it.
If you run BeyondTrust Remote Support yourself, make sure update BT26-05 is installed. BeyondTrust released it on October 8, and it fixes the zero-day this intrusion started with.
#Is your vCenter affected by CVE-2026-123890?
Any vCenter older than these builds is affected, including servers you patched for July's flaw:
| System | Status |
|---|---|
| vCenter 9.1 (VCF and vSphere Foundation 9.1) | Fixed in 9.1.1.0200 |
| vCenter 9.0 (VCF and vSphere Foundation 9.0) | Fixed in 9.0.2.0200 |
| vCenter 8.0 | Fixed in 8.0 U3m |
| VMware Cloud Foundation 5.x | Fixed in 5.2.5 |
| Telco Cloud Platform and Telco Cloud Infrastructure | Fixed in 8.0 U3m (Broadcom KB463116) |
| vCenter 7.0 | No public patch: extended-support customers contact Broadcom |
The vSphere Client shows vCenter's version and build on its Summary tab. From PowerCLI it looks like this (we don't have a vCenter to run it against, so there's no output shown):
student@opnf:~$ Connect-VIServer vcenter.example.com student@opnf:~$ $global:DefaultVIServer.Version student@opnf:~$ $global:DefaultVIServer.Build # 8.0 U3m is version 8.0.3.01200, build 25934264
Then check who can reach port 2012, the port the exploit talks to. Swap localhost for your vCenter's address and run it from each network segment. Here it's pointed at a machine with nothing listening, so you can see what a blocked or closed port looks like:
student@opnf:~$ nc -zv -w 3 localhost 2012 nc: connect to localhost (127.0.0.1) port 2012 (tcp) failed: Connection refused # "succeeded" instead means this segment can reach vCenter's SSO interface
Cutting down which machines can reach vCenter shrinks the number of places an attack can start from. It doesn't replace the update, though, as Cyber Security News points out: in Google's case, the attack came from a box already inside the network.
How to patch it
Broadcom calls this an emergency change. Three things to know before you start:
- Use the full updateThese fixes can't be installed through vCenter Quick Patch, so plan for the regular update process.
- Expect a short management outageVirtual machines keep running, but vCenter's management access drops while it updates, and some encryption operations that use a Standard Key Provider wait until it's back.
- Check your VCF 9.1.1 plans firstInstalling vCenter 8.0 U3m or 9.0.2.0200 blocks a later upgrade to VCF 9.1.1. Either upgrade vCenter to 9.1.1 first and patch there, or patch now and wait for VCF 9.1.2.
Signs you were already hit
Patching closes the hole but doesn't remove anyone who got in first. Mandiant's CTO, Charles Carmakal, said organizations should "both apply the patches and examine their systems for evidence of compromise." Based on what the attackers did in Google's case, look for:
- Single sign-on administrator accounts nobody on your team created (in the vSphere Client, under Administration, then Single Sign On, then Users and Groups).
- New root SSH keys on your ESXi hosts.
- Unusual access to virtual disks, especially the disks of domain controllers.
- Linux appliances and remote-access boxes, such as BeyondTrust Remote Support, that can reach vCenter with no EDR watching them.
Google's report lists the malware involved, including BONESAW and SAWSNAKE, along with indicators you can load into your security tools. If you find any of these signs, bring in incident response: with the SAML signing material stolen, the attackers can forge logins without a password, so password resets alone won't lock them out.
If you can't tell either way, patch first, then treat any vCenter that an unmonitored or compromised machine could reach on port 2012 as suspect until you've checked it.
#Broadcom's advisory doesn't mention the attacks
If you only read Broadcom's advisory, VMSA-2026-0009, you'd see a critical bug with a patch, not one that's been used in attacks for months. As SecurityWeek notes, Broadcom credits Mandiant with reporting it but says nothing about exploitation. CISA hadn't added it to its Known Exploited Vulnerabilities catalog when Rescana checked on the afternoon of November 19, and Rescana warns that a listing could follow Google's report. Here, Google's write-up is the one that tells you how urgent it is.
The same advisory fixes four lower-severity flaws in SDDC Manager and VMware Tools for Windows, rated 6.7 to 8.1. Each needs an attacker who's already signed in or privileged.
Working out what an attacker did after they got in, and what they left behind, is incident response work. The Security Pro ProCourse covers it in its Incident Response & Forensics level, hands-on in a simulated terminal in your browser.
Does patching ESXi fix CVE-2026-123890?
Does it matter if I use Enhanced Linked Mode?
Are vSphere 6.5 and 6.7 affected?
What this means for you
Patch vCenter as an emergency change, check the single sign-on administrators for accounts nobody created, and find every Linux appliance or remote-access box that can reach vCenter without EDR watching it.
SourcesRapid7SecurityWeek, citing GoogleGoogle Threat IntelligenceThe Hacker NewsCyber Security NewsBeyondTrustRescanaSecurity Affairs, citing CrowdStrike
Get opnF in your inbox
Tech news for IT professionals that actually teaches you something.
Our promise to you: we'll be conscious about email quality/frequency, and your address will never be shared/sold.