China’s ‘Armory Castle’ Lays Siege to VMware vCenter Zero-Day
During the last wave of major VMware vCenter security updates, a China-nexus threat group likely already had its next zero-day up its sleeve.
Chinese state-sponsored hackers have been using a newly disclosed VMware zero-day vulnerability and an array of custom malware to compromise victims’ virtualization stacks for several months.
In October, Mandiant described a campaign by China’s Armory Castle (formerly UNC5221) to exploit organizations through BeyondTrust’s Remote Support tool. The group used a zero-day to breach the edge appliance, then stole credentials and secrets useful for network reconnaissance and downstream compromises within organizations. Exactly what that downstream compromise looked like, though, was unclear at the time.
It appears now that BeyondTrust was merely an entrée to a deeper, more interesting, and much more complex campaign. In a follow-up blog post this week, Mandiant researchers described a chained attack beginning with Beyond Trust, moving through obscure Linux appliances, and landing on VMware technologies, at which point the attackers utilized another zero-day and more than half a dozen malware programs to totally undress their targets’ virtual machine (VM) environments.
That zero-day has only now come to light, months since China first exploited it. In a security advisory, VMware owner Broadcom recommended that customers update their systems. Tracked as CVE-2026-123890, the vulnerability allows for authentication bypass in VMware vCenter, and earned a “critical” 9.8 out of 10 score in the Common Vulnerability Scoring System (CVSS).
Armory Castle Exploits VMware vCenter Flaw
From tagged Remote Support boxes, Mandiant observed Armory Castle scanning a victim’s internal environment, looking for open ports associated with VMware.
Instead of going for the goods immediately, the group established forward operating bases inside of other, possibly unrelated Linux appliances in the same network. These appliances lacked endpoint detection and response (EDR) visibility, allowing the group to establish persistence and stage its malware arsenal away from the parts of the network where defenders were looking. Among other malware, it was there that Armory Castle deployed “Bonesaw,” a Go-based remote exploitation tool customized for vCenter.
The VMware vCenter Server is the central control plane used to manage multiple VMs and their ESXi hosts. Because it oversees everything else, few technologies are more useful for lateral movement, ransomware attacks, etc. than vCenter is. It’s why this summer, the disclosure of two critical vCenter vulnerabilities, CVE-2026-59309 and CVE-2026-59310, caused a stir.
While organizations were rushing to patch the two vCenter vulnerabilities known to Broadcom, Armory Castle appears to have had an extra one up its sleeve. In September, it was able to use Bonesaw against a victim that had already patched its vCenter; Bonesaw was tuned to exploit CVE-2026-123890, a flaw that stems from the vCenter login system’s use of predictable, time-based UUIDv1 timestamp generation in Secure Remote Password (SRP) authentication handling.
Exploiting CVE-2026-123890 afforded Armory Castle a shared session key it could use to create new authenticated sessions as an administrator, bypassing the normal authentication process entirely. Bonesaw then paid more dividends, automating the process of copying vCenter’s internal directory database, and creating powerful new accounts for further damage.
At this point, Armory Castle had total control over the victim’s vCenter. Yet it continued on with machine-like focus; the attackers deployed a tool called “Sawsnake” to turn the Bonesaw-stolen vCenter login-signing key into forged authentication tokens, bypassing multifactor authentication (MFA) and enabling impersonation of any user. It then jumped to ESXi hypervisors, and found and extracted system files.
Protecting VMs from Nation-State Threats
For high-risk organizations, updating VMware instances to patched versions can’t be the end of the story.
Because vCenter is just so sensitive, Mandiant urged organizations not to rely solely on patching known vulnerabilities as they come up. It pointed to its vSphere Defender’s Guide, published earlier this spring. Its recommendations include the standard fare — promptly updating and isolating management systems, encrypting particularly sensitive systems, and applying especially tight controls to administrator accounts — but also offers more pointed items. It notes that vCenter doesn’t necessarily record all important operating system (OS) activity, and that sophisticated attackers can delete local logs post-compromise; therefore, organizations should consider forwarding logs to a separate security information and event management (SIEM) platform.
Mandiant researchers also advised that organizations not rely on vCenter’s built-in firewall, as it has too many limitations. For example, it doesn’t generate remote logs for rejected connection attempts or certain shell activity, doesn’t enforce policies for data egress, and, most of all, it’s simply at risk for being built into the very system it’s designed to secure.
“Organizations must adopt a proactive, Zero Trust approach to infrastructure hardening to defend against current and undiscovered zero-day vulnerabilities,” the researchers wrote.
Originally published on Dark Reading