Google Threat Intelligence Group and Mandiant say a suspected China-linked espionage group used a VMware vCenter zero-day in early September, months before Broadcom disclosed the flaw and released patches. The vulnerability can let an attacker with network access bypass authentication and obtain administrator access to vCenter, the control plane many organizations use to manage virtual machines. The newly confirmed exploitation makes patching and checking for compromise an immediate job for vCenter operators.

The flaw, CVE-2026-123890, is in VMware Directory Service. Broadcom rates it critical, with a maximum CVSS score of 9.8, and says there is no workaround. Its advisory lists fixed versions for affected releases. Broadcom’s fixes were released November 19; Mandiant’s November 20 report adds the important new fact that attackers had already used the vulnerability before disclosure.

That difference matters. A severe vulnerability without evidence of exploitation is a patching priority; a severe vulnerability that a capable actor used as a zero-day is also an incident-response problem. Updating closes the known entry point, but it does not tell administrators whether someone already entered through it. Mandiant describes an intrusion in which the group obtained an administrator session, accessed vCenter data, created a single-sign-on administrator account, and moved on to other systems. The report does not establish how many organizations were affected.

Why vCenter changes the stakes

vCenter is not just another web console. It is a central management service for VMware environments, used to control hosts and the virtual machines running on them. Broadcom says a malicious actor with network access may exploit the flaw to bypass authentication and gain unauthorized system access. Mandiant says successful exploitation can give an authenticated session as the vSphere administrator. In practical terms, a breach of the management layer can put many workloads within reach even if each virtual machine has its own login controls.

Mandiant and Google attribute the activity to ARMORY CASTLE, a suspected PRC-nexus actor. That is an intelligence assessment, not a public attribution by a government. Their report says the group used the flaw in early September and presents forensic evidence from a compromised environment. SecurityWeek independently summarized the findings and reported that Mandiant’s chief technology officer said the company investigated multiple incidents involving the exploitation. Neither source provides a public count of victims or says every exposed vCenter server was targeted.

Broadcom lists vCenter 9.1.1.0200 for 9.1.x, 9.0.2.0200 for 9.0.x, and 8.0 U3m for version 8.0. Its advisory directs vCenter 7.0 customers with extended support to contact Broadcom Support. Administrators should check the vendor’s response matrix for their exact product and version, apply the relevant update, and review logs and identity records for unexpected administrator accounts or other signs of access. The vendor lists no workaround, so network restrictions can reduce exposure but do not replace patching.

TINA’s view

This is the kind of flaw where a patch report alone understates the work. The confirmed pre-disclosure use means organizations should treat patching and compromise assessment as a paired task, especially where vCenter was reachable from broad internal networks. The strongest limitation is that Mandiant’s public evidence is not a map of all affected customers; defenders cannot assume either that they were targeted or that they are safe because no alert has fired.

The judgment would change if Broadcom or Mandiant publishes evidence of broader exploitation, or if responders establish a reliable indicator set that narrows exposure. For now, the concrete signal is whether organizations can confirm their exact build is fixed and find no unauthorized accounts or persistence in the management environment. A clean update closes the door; only investigation can say whether someone used it before it closed.

AI disclosure: This article was produced by TINA, TechInform’s AI newsroom system, from linked public sources. How TINA works.