CyberSecurityNews

Hackers Exploit VMware vCenter 0-Day Since September to Hijack Admin Accounts and Steal VM Data


A suspected China-linked espionage group exploited a VMware vCenter Server zero-day in early September 2026, gaining administrator access and stealing files from a Windows virtual machine.

Mandiant and Google Threat Intelligence Group (GTIG) attributed the intrusion to ARMORY CASTLE, formerly tracked as UNC5221. Broadcom disclosed the flaw, CVE-2026-123890, on November 19, more than two months after the earliest use documented in the investigation.

The flaw carries a severity score of 9.8 out of 10. It affects VMware Directory Service, which supports vCenter’s single sign-on system. An attacker who can reach a vulnerable server over the network can bypass authentication and open a session as the built-in vSphere administrator without knowing the account’s password. In the incident Mandiant investigated, the attackers reached vCenter from inside the network.

That distinction matters because vCenter is the central management system for VMware vSphere. ESXi hosts run the virtual machines, while vCenter lets administrators control those hosts and the workloads they carry.

A break-in at vCenter can therefore put many systems at risk, even when the server sits on a private network, and the virtual machines have their own security tools.

The intrusion began with a separate zero-day, CVE-2026-102764, in BeyondTrust Remote Support. Mandiant traced that first breach to late August. The attackers then moved to an internal Linux appliance that lacked endpoint detection and response software.

VMware vCenter 0-Day (Source: Google)

They gained root access and installed tools that let them keep access and look for VMware systems. The vCenter flaw helped them extend an existing breach rather than start one.

ARMORY CASTLE Took Over vCenter

Our earlier coverage of the BeyondTrust Remote Support zero-day explains how the group gained its first foothold. The Mandiant and GTIG vSphere investigation picks up from there.

Investigators found that the group used a custom tool called BONESAW against a vCenter server that had already been patched for CVE-2026-59309, a separate authentication bypass disclosed in July. That earlier update did not block the new attack.

BONESAW let the group copy the vCenter directory database and create a new single sign-on administrator account. The attackers also deployed SAWSNAKE, a tool that could use stolen signing keys to forge SAML tokens.

These tokens tell vCenter that a user has already been authenticated. If attackers steal the signing keys, they can impersonate an administrator without going through the normal sign-in process or its multifactor checks.

The attackers stole vCenter encryption material and database credentials before reaching an ESXi host. There, investigators saw them access files stored inside a Windows virtual disk, including Active Directory data, registry files, and DNS data.

The group copied an archive off the ESXi host and moved the data back through the compromised Linux appliance. Mandiant assessed that the attackers likely sent it out through the BeyondTrust appliance.

The theft illustrates why monitoring Windows alone is not enough. When someone reads a virtual disk from the host, the activity takes place outside the guest operating system, where its endpoint tools may not see it. Our coverage of BRICKSTORM attacks on VMware ESXi and Windows describes a separate campaign that also targeted VMware infrastructure to reach sensitive data.

VMware vCenter Patches

Broadcom’s security advisory VMSA-2026-0009 lists fixes in vCenter 9.1.1.0200, 9.0.2.0200 and 8.0 Update 3m. VMware Cloud Foundation 5.x customers should follow Broadcom’s guidance for 5.2.5.

Broadcom lists no public fix for vCenter 7.0 and directs customers with extended support contracts to contact support. There is no vendor workaround. Restricting access to management networks reduces exposure but does not replace the update.

Patching closes the flaw but does not remove a rogue administrator account or make a stolen signing key safe again. Security teams should preserve logs from vCenter, ESXi and connected appliances, then review them for unexpected administrator accounts, brief periods when SSH was enabled, new ESXi root SSH keys and unusual access to virtual disks.

Correlating those events across systems is especially important when a remote support appliance can reach the VMware management network.

If investigators find signs of compromise, the response should extend beyond VMware. Teams should revoke affected sessions, replace exposed signing keys and credentials, and assess whether stolen Active Directory data puts the wider Windows domain at risk.

The key question is not only whether vCenter is patched today, but whether someone used it to reach systems and data before the fix arrived.

What Are Leading Organizations Doing About AI and Cyber Risks? Find Out. View Details!



Source link