GhostFrame Malware 2026: New VMware ESXi Backdoor Linked to Settra Ransomware Attacks

Recorded Future’s Insikt Group has documented GhostFrame, a previously unreported post-exploitation framework that gives attackers a single control panel over compromised VMware ESXi hypervisors. The tool was found in infrastructure tied to TAG-218, a cluster whose activity overlaps with Settra ransomware, and it matters well beyond IT: plants that run historians, engineering workstations and SCADA servers as virtual machines sit on exactly this layer.

Insikt Group published the research on November 10, 2026. This article separates what the research establishes from what remains unknown, because several early summaries have blurred that line.

What was discovered

GhostFrame is a command-and-control (C2) and post-exploitation framework built for ESXi. It is not an exploit. Per Insikt Group, an operator must already hold root access to the ESXi host before deploying it. Once there, the framework lets the operator maintain access, run commands, manage virtual machines, pivot into internal networks and harvest credentials, including from Windows guests, from one web-based interface.

Insikt Group found it while monitoring operator infrastructure behind TAG-218. Between late August and October 2026, a cluster of TAG-218 victim-facing servers targeted VPN appliances in the US and Europe, predominantly SonicWall, Palo Alto Networks and Fortinet devices. Four of the targeted organizations were later listed on the Settra ransomware leak site. The victims have not been named publicly, and their sectors have not been disclosed.

On October 12, 2026, a victim Fortinet VPN appliance was observed beaconing to both a GhostFrame C2 server and a MeshCentral remote-management server that Huntress had previously tied to Settra intrusions. That observation is what places GhostFrame inside the TAG-218 intrusion chain.

How GhostFrame works

Implant, persistence and defense impairment

The operator panel has two sections, “Sessions” and “Builder.” The builder generates an ELF implant named vmhostd-core, and its instructions tell operators to compile it, manually disable the ESXi firewall and run it from a datastore.

In the analyzed sample, the implant:

  • copies itself to /opt/vmhostd-core;
  • modifies the ESXi boot script /etc/rc.local.d/local.sh so it launches at startup;
  • repeatedly disables the ESXi firewall and sets execInstalledOnly to 0 (the control that restricts execution of unsigned binaries), roughly every five seconds, so an administrator who re-enables them is quickly overridden;
  • runs /sbin/auto-backup.sh so the modified configuration survives reboots;
  • changes its process name and command line to vmhostd, resembling VMware’s legitimate vmware-hostd daemon.

C2 behavior and in-memory execution

The sample tries simultaneous TCP connections to a single C2 address across six ports: 53, 2222, 3389, 4444, 8443 and 9443. It keeps the first connection that succeeds and closes the rest. If nothing connects within 30 seconds, it retries after a random 5 to 12 second delay.

Once connected, it acts mainly as an in-memory shellcode loader: it receives machine code from the server, allocates executable memory and runs it without writing the payload to disk. If memory allocation fails, it falls back to a standard /bin/sh reverse shell.

Post-exploitation capabilities

CapabilityWhat Insikt Group reports
VM managementEnumerate VMs and power states, change power state, create and delete snapshots
Guest accessEnable VNC on individual VMs and relay graphical sessions through the C2 to the operator panel
Management accessExpose the ESXi HTML Host Client through an operator-controlled tunnel
Network pivotinggfsox, a SOCKS5 proxy deployed on the host to reach internal systems
Credential interceptionsdkmon, described in the frontend as modifying ESXi service configuration to capture usernames and passwords from later ESXi logins
Guest credential theftIntegrated open-source VMkatz, working from VM memory snapshots, virtual disk images, both, or a recursive scan of a datastore; targets include NTLM hashes, WDigest, Kerberos tickets, SAM and LSA secrets, and NTDS.dit

The practical consequence, in Insikt Group’s framing: credentials can be pulled from hosted Windows systems without running credential-stealing malware inside each guest. Anything that relies only on in-guest EDR visibility will not see that activity. That inference is ours, but it follows directly from where the collection happens.

Root cause and the VPN-to-hypervisor path

GhostFrame itself has no initial-access capability, so the real question is how attackers reached root on ESXi. The research does not state how the four leak-site victims were compromised. What it does say is that TAG-218 infrastructure was aimed at VPN appliances, and that this aligns with recent reporting on CVE-2026-0257, an authentication bypass in Palo Alto Networks GlobalProtect.

Context on that flaw, from Unit 42 and other published reporting:

  • It lets an unauthenticated attacker forge a valid authentication override cookie and establish a VPN session without credentials or MFA. Published scores are CVSS 3.1 of 9.1 and CVSS 4.0 of 7.8.
  • Palo Alto Networks published its advisory on May 13, 2026. Rapid7 reported first observing exploitation on May 17. CISA added it to the Known Exploited Vulnerabilities catalog on May 29.
  • Unit 42’s October 19 update says ransomware operators, including Qilin and Inc, have attempted exploitation, and that public reporting attributes exploitation to actors using Settra ransomware.
  • Unit 42 calls the most significant evolution in post-exploitation tradecraft the direct pivot to the virtualization layer: disabling ESXi execInstalledOnly and using a custom tool for offline credential dumping against a domain controller’s virtual disk (VMDK).
  • Recommended mitigation includes applying the fixed PAN-OS release and disabling the Authentication Override options in GlobalProtect portal and gateway configuration.

What is not established: Recorded Future says the activity is consistent with this reporting. It does not say that every TAG-218 victim was compromised through CVE-2026-0257, and it also names SonicWall and Fortinet targets. Treat the VPN link as a well-supported pattern, not a proven per-victim root cause.

Attribution: what is confirmed and what is not

  • Supported by evidence: TAG-218 infrastructure connected to two GhostFrame C2 servers and a MeshCentral server previously associated with Settra; four targeted organizations later appeared on Settra’s leak site.
  • Stated limits: Insikt Group describes the relationship as an overlap. Available evidence does not show GhostFrame is exclusive to Settra, and its distribution model is unknown. Some news coverage describes GhostFrame as “used by Settra,” which is stronger than the source.
  • About Settra: It surfaced in June 2026 as a ransomware and data-extortion group. Huntress, which investigated two incidents (July and September), notes that nothing published establishes a ransomware-as-a-service model. Its observed Windows playbook includes MeshAgent RMM for persistence, the gdrv.sys vulnerable driver in one case, log clearing, and recovery inhibition via reagentc /disable. Ransom notes are named RESTORE_FILES.txt. SOCRadar’s tracker lists well over 90 claimed victims, but leak-site claims are not verified breaches.
  • Unknown: The identity, location or motive beyond financial gain of any individuals. No state sponsorship is alleged in the research.
  • Forward-looking assessment: Insikt Group rates it highly likely that other financially motivated actors, including ransomware affiliates and initial access brokers, will adopt GhostFrame or similar ESXi tooling.

Impact assessment

No victim-level impact figures, ransom demands, downtime or regulatory actions have been published for the four leak-site organizations, and we are not estimating any. What can be said:

  • Technical: A root-level hypervisor foothold exposes every VM on the host and collapses the isolation many teams assume between guests. Credential theft via snapshots and VMDKs can feed domain-wide compromise.
  • Operational: Snapshot and power-state control means an attacker with this access could disrupt or encrypt workloads at the infrastructure level. Whether GhostFrame was used to deploy encryption in any specific incident is not stated in the research.
  • Detection: Persistence that keeps re-disabling security controls, plus in-memory payloads, makes simple re-enabling of settings insufficient. Insikt Group advises finding and removing the underlying persistence instead.

Why this matters for OT, ICS and critical infrastructure

Many industrial sites virtualize supporting systems on ESXi: historians, engineering workstations, HMI and SCADA servers, patch and AD servers, jump hosts. Where that is the case, a compromised hypervisor puts those systems in reach of an attacker who may never touch an IT-side guest agent.

Be precise about the limits. The research establishes no PLC or controller compromise and no OT-specific targeting. The relevance is architectural: Dragos’s Q2 2026 analysis counted 1,140 ransomware incidents affecting industrial organizations (747, or 65%, in manufacturing) and found no case where a ransomware operator reached Stage 2 of the ICS Cyber Kill Chain or directly manipulated a control system. Disruption came from losing IT and supporting systems. A hypervisor is a high-leverage place to cause that loss.

Industrial teams should ask: which ESXi hosts carry OT-adjacent VMs, how is their management plane reached, and does a VPN appliance sit on that path?

Incident response and recovery status

There is no victim-side response narrative to report. We found no public statement from Broadcom (VMware’s owner) on GhostFrame at the time of writing, and no dedicated government advisory for it. Insikt Group’s own mitigation guidance is the closest thing to an official response. Because the tool requires root and not a VMware flaw, a vendor patch would not address it directly.

Detection and hunting guidance

From Recorded Future’s report:

  • Check /etc/rc.local.d/local.sh for unexpected changes.
  • Look for /opt/vmhostd-core and a process named vmhostd.
  • Alert on firewall disablement and execInstalledOnly set to 0, and investigate the persistence, not just the setting.
  • Alert on a burst of about six near-simultaneous TCP SYNs from an ESXi host to one external IP across ports 53, 2222, 3389, 4444, 8443 and 9443. This behavior can survive C2 rotation.

Indicators of compromise (from Recorded Future; defanged)

IndicatorDescription
193[.]5[.]65[.]114TAG-218 MeshCentral C2
45[.]158[.]196[.]81TAG-218 GhostFrame C2
45[.]158[.]196[.]141TAG-218 GhostFrame C2
a05eae6d07b82e2e5451a72266f21fbaf76486ef4920a820988de93e15353a10GhostFrame SHA-256

Note: one outlet claimed the published hash is truncated. We checked it: it is 64 hexadecimal characters, a complete SHA-256. The address 193.5.65[.]114 also appears in Huntress’s September Settra incident as a MeshAgent C2.

Security recommendations

Items 1 to 3 follow Insikt Group’s mitigations; the rest are our general hardening suggestions, not claims from the research.

  1. Restrict and monitor root access to ESXi. Limit management access to trusted systems, and investigate unexpected privileged logins or host changes.
  2. Hunt persistence, not settings. Treat any repeated firewall or execInstalledOnly changes as a compromise indicator.
  3. Watch egress from hypervisors. ESXi hosts rarely need arbitrary internet access; the six-port burst is a strong signal.
  4. Patch and harden VPN edges. Apply fixed PAN-OS versions for CVE-2026-0257, disable Authentication Override if not needed, and review SonicWall and Fortinet appliances for current advisories. Review logs for unauthorized sessions.
  5. Separate the management plane. Keep ESXi and vCenter management on an isolated network reachable only through MFA-protected, PAM-controlled jump hosts; do not allow VPN users direct reach.
  6. Use platform protections. Keep execInstalledOnly enforced, consider lockdown mode and secure boot with TPM-backed attestation per Broadcom’s hardening guidance, and disable unused shells and SSH.
  7. Protect domain controllers and snapshots. Limit who can create, export or copy snapshots and VMDKs; treat DC virtual disks as crown-jewel data and monitor datastore access.
  8. Plan recovery for the hypervisor layer. Keep offline, immutable backups and a rebuild plan for ESXi hosts, since a compromised host should be reimaged.
  9. OT segmentation. Ensure virtualized OT-adjacent workloads have a documented dependency map, so hypervisor loss does not unexpectedly stop production.

Leave a Reply

Your email address will not be published. Required fields are marked *