China-Linked Hackers Exploited a Critical VMware vCenter Flaw for Months Before the Patch

A suspected China-linked espionage group known as Armory Castle exploited a critical VMware vCenter authentication bypass, CVE-2026-123890, as a zero-day since early September 2026. Google Threat Intelligence Group and Mandiant revealed the campaign on 20 November, a day after Broadcom patched the flaw. The vulnerability carries a CVSS score of 9.8, has no workaround, and lets an unauthenticated attacker seize full administrator control of vCenter.

Nov 22, 2026 - 08:38
Nov 22, 2026 - 08:39
 0  2
China-Linked Hackers Exploited a Critical VMware vCenter Flaw for Months Before the Patch
Server racks with VMware VCF hardware and blue LED lighting in a modern data center.

China-Linked Hackers Exploited a Critical VMware vCenter Flaw for Months Before the Patch

A suspected China-linked cyber espionage group quietly exploited a critical flaw in VMware vCenter Server for more than two months before a patch existed, using it to take full administrative control of virtual infrastructure and steal sensitive data. Google Threat Intelligence Group and Mandiant disclosed the campaign on 20 November 2026, one day after Broadcom shipped the fix.

The group, tracked as Armory Castle and formerly known as UNC5221, turned the vulnerability into the centrepiece of a wide-ranging intrusion that also involved a second zero-day, a set of custom-built hacking tools and a bespoke malware toolkit designed specifically for VMware environments.

A flaw that hands attackers the keys to vCenter

The vulnerability is tracked as CVE-2026-123890 and carries a CVSS score of 9.8, the highest severity rating. It is an unauthenticated Secure Remote Password (SRP-6a) session key disclosure flaw in the VMware Directory Service, known as vmdird, reachable over Distributed Computing Environment and Remote Procedure Calls (DCE/RPC).

According to Google Threat Intelligence Group and Mandiant, successful exploitation produces a remote authentication bypass that creates an authenticated session as administrator@vsphere.local — the built-in vCenter administrator — without any password. From there, an attacker can run arbitrary commands on the appliance and pull out files, including the vCenter directory database, data.mdb.

vCenter is the central control plane for VMware virtual environments, managing ESXi hosts and the virtual machines running on them. Whoever controls vCenter effectively controls the whole estate, which is what makes this flaw so dangerous. There is no workaround: patching is the only fix.

The attack began somewhere else entirely

The vCenter exploitation was not the start of the intrusion. Google's researchers say the campaign began with the exploitation of a then zero-day flaw in BeyondTrust Remote Support, tracked as CVE-2026-102764 with a CVSS score of 9.3, which the group used for reconnaissance, to gain administrative control and to establish persistence. It also deployed a custom proxy tunneller called BARECLAW.

From the Remote Support device, the attackers moved laterally to an internal Linux appliance, escalated their privileges to root and deployed further payloads to map the environment, steal data and push into the virtualised infrastructure. They used the network scanner nmap to look for systems running VMware-specific ports — 9080, 5480 and 2012. Port 2012 is the control interface RPC for vCenter Single Sign-On, and it proved central to the exploitation that followed.

BONESAW and a toolkit built for VMware

The centrepiece of the vCenter attack is a utility the researchers call BONESAW. Written in the Go programming language, it is built specifically to exploit CVE-2026-123890 remotely. It exfiltrates the vCenter Lightning Memory-Mapped Database (LMDB), hijacks the SAML Security Token Service and creates a Single Sign-On administrator account.

Google identified six malware families and utilities deployed during the campaign. Alongside BONESAW there were SLIMSAVAGE and CAGEMATCH, JSP web shells used to run arbitrary commands, stage payloads and hold a foothold on the Linux appliance; SAWSNAKE, a Go utility that enables vCenter SAML authentication bypass and session cookie generation, making Golden SAML attacks possible so the attackers could forge authentication tokens, bypass multi-factor authentication and impersonate any user; BEARBONES, a C backdoor based on the open-source Dropbear SSH server that can bypass system password authentication and log plaintext credentials; and LAZYBONES, a lightweight Go backdoor that uses Slack for command-and-control communications.

The researchers note that BONESAW represents a significant expansion into post-compromise use of zero-days. The group's use of VMware-specific tools such as SAWSNAKE alongside malware tailored to Remote Support and Linux systems shows, in Google's words, a deep understanding of each appliance it targeted. Using three separate specialised sets of malware and two zero-days in a single campaign, Google said, indicates significant resources behind Armory Castle and the extent of its commitment to compromising high-priority targets.

Inside the compromised vCenter

Once inside vCenter, the attackers created a local administrator account named sps-@vsphere.local and enabled SSH on the appliance through the vCenter Server Appliance Management Interface (VAMI) API, sending a PUT request to /rest/appliance/access/ssh. They then connected over SSH using the newly created account.

In the final stage, they enumerated the ESXi file system, mounted Virtual Machine File System (VMFS) volumes and added an SSH public key to the root account's authorised keys. They also carried out further reconnaissance and data theft on the ESXi file system, archiving the stolen data and downloading it through HTTP GET requests to the Remote Support appliance for exfiltration. LAZYBONES was also dropped on a separate Linux appliance, where it periodically checked a designated Slack channel for commands, allowing the group to run commands, transfer files and adjust its beacon interval.

It is suspected that Armory Castle knew about the flaw before a related vCenter authentication bypass, CVE-2026-59309, was disclosed on 29 July 2026. The recovered BONESAW binary carries a build timestamp of 13 July 2026, although Google notes that build metadata can be modified before compilation. Exploitation of CVE-2026-123890 itself began in early September 2026, more than two months before the patch.

Which versions are fixed

Broadcom addressed the flaw on 19 November 2026 in advisory VMSA-2026-0009. The fixed versions are VMware Cloud Foundation and VMware vSphere Foundation 9.1 in release 9.1.1.0200, the 9.0 line in 9.0.2.0200, VMware vCenter 8.0 in update 8.0 U3m, and VMware Cloud Foundation 5.2.5. The Telco Cloud Platform and Telco Cloud Infrastructure products are fixed through vCenter 8.0 U3m. Customers still on vCenter 7.0 should contact Broadcom Support if they hold an extended support contract.

Broadcom's own advisory does not itself confirm in-the-wild exploitation, but the US Cybersecurity and Infrastructure Security Agency added CVE-2026-123890 to its Known Exploited Vulnerabilities catalogue on Friday 20 November. Google's attribution of the exploitation to Armory Castle rests on its incident-response investigation and the recovered tooling.

The flaw is the second VMware vulnerability Armory Castle has exploited as a zero-day this year. Late in October, CrowdStrike reported incidents in which the group abused CVE-2026-41709, a low-severity insufficient logging flaw in VMware ESX rated 2.7, to reduce audit trail visibility.

What administrators should do now

Organisations running vCenter should apply the patched releases immediately, since there is no workaround and the flaw is known to have been exploited in the wild. Because the attackers established persistence, created rogue administrator accounts and stole directory data, patching alone is not enough: administrators should also review vCenter for unexpected administrator accounts, check for unauthorised SSH access and changes to SSH settings, monitor access to vCenter management services — especially TCP port 2012 — and look for suspicious single sign-on account creation, unexpected administrative sessions and unusual use of the VAMI API. Any vCenter that may have been exposed should be treated as potentially compromised and investigated accordingly.

Sources

What's Your Reaction?

Like Like 0
Dislike Dislike 0
Love Love 0
Funny Funny 0
Angry Angry 0
Sad Sad 0
Wow Wow 0