Image: storage.googleapis.com · rights & removal
ARMORY CASTLE Exploits vSphere
Reporting by Google Cloud Threat IntelligenceRead the original at cloud.google.com
Executive Summary
Facts Only
* ARMORY CASTLE exploited CVE-2026-102764 in BeyondTrust Remote Support.
* ARMORY CASTLE exploited CVE-2026-123890 in VMware vCenter Server, which allows for unauthenticated Secure Remote Password session key disclosure over DCE/RPC.
* The exploitation of CVE-2026-123890 via BONESAW enabled remote authentication bypass and access to the data.mdb database by leveraging the vCenter Server Directory Service RPC interface.
* ARMORY CASTLE used BONESAW to exfiltrate LMDB and hijack SAML STS to create an administrator@vsphere.local account.
* Lateral movement involved scanning internal Linux appliances for VMware-specific ports (9080, 5480, 2012).
* ARMORY CASTLE deployed five malware families: SLIMSAVAGE, CAGEMATCH webshells, BEARBONES backdoor, BONESAW utility, and SAWSNAKE SAML utility.
* SLIMSAVAGE acted as a command execution webshell utilizing ProcessBuilder to execute commands via /bin/sh.
* CAGEMATCH functioned as a payload staging webshell, extracting file contents from the `/tmp` directory.
* Privilege escalation to root on Linux appliances was achieved by enabling SUID permissions on /bin/env using gdb.
* BONESAW enabled exfiltration of data.mdb and RSA private keys.
* SAWSNAKE enabled vCenter SAML authentication bypass and session cookie generation via data.mdb parsing.
* BEARBONES backdoor was deployed to facilitate SSH access, including setting up a listener on TCP ports 53 and 33992.
* ARMORY CASTLE exfiltrated ESXi filesystem data from guest VMs using SCP commands against the ESXi host.
Full Take
The observed sequence of activity reveals a sophisticated, multi-stage post-compromise methodology by ARMORY CASTLE that moves beyond simple exploitation into deep environment compromise and persistence. The use of BONESAW to exploit vCenter’s underlying RPC flaw to achieve authentication bypass is a critical demonstration of exploiting platform trust boundaries for data exfiltration. The subsequent lateral movement and deployment of custom, layered malware—SLIMSAVAGE for command execution, CAGEMATCH for staging, and SAWSNAKE for SAML manipulation—suggest an actor with deep, specific knowledge of the VMware ecosystem and Linux operational details. This layering indicates a resource-backed operation targeting high-value virtualization infrastructure.
The progression from initial zero-day exploitation to leveraging platform-specific tools like BONESAW and SAWSNAKE suggests an intent not merely to gain initial access, but to establish persistent, privileged control over the virtual environment itself. The final phase, where BEARBONES was used to pivot to SSH on vCenter via VAMI manipulation, culminating in credential theft from master keys and PostgreSQL data, highlights a strategy focused on controlling the entire hypervisor stack. This operational complexity implies that defense strategies must move beyond patching individual vulnerabilities to implementing infrastructure-centric controls, such as those recommended for hardening the vCenter appliance itself with kernel-level filtering rather than relying solely on application interfaces. The implication is that centralized control plane security requires enforcing Zero Trust principles across management interfaces, preventing internal lateral movement even after an initial foothold is achieved.
Bridge Questions:
What other vulnerabilities within VMware or related infrastructure might be chained together using similar RPC/RPC-related flaws to achieve session key disclosure or authentication bypass? How effectively do existing vulnerability management processes account for the dynamic interaction between zero-day flaws and custom exploitation tools like BONESAW when assessing overall risk to critical infrastructure? Does relying on application-level security features, such as host firewalls via VAMI, adequately mitigate risks stemming from kernel-level privilege escalation paths exploited by post-exploitation frameworks?
From the original · Google Cloud Threat Intelligence
Google Threat Intelligence Group Mandiant Written by: Jayson Frost, Bryce Greathouse, Dhanesh Kizhakkinan, John Palmisano, Steve Rasch, Daniel Sislo, John Wolfram Introduction On Oct.Read the full story at cloud.google.com
Sentinel unavailable
The automated check of the source article's wording did not complete, so no result is shown.
This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.
