Back to Articles

SOC

Hypervisor ransomware: vCenter, ESXi and the single point that stops a whole company

12 min read

Why ransomware groups encrypt VMware ESXi instead of individual servers, what the August 2026 vCenter vulnerability (CVE-2026-59310) taught us, and how to isolate management, protect identities and backups, and spot the warning signs.

ransomware
esxi

On 18 August 2026 CISA added CVE-2026-59310, a path traversal in VMware vCenter, to its Known Exploited Vulnerabilities (KEV) catalog and marked it as used in ransomware campaigns. Three months later, with the full picture available, the lesson is clear: for a large share of European mid-market companies the most valuable target on the network is not a file server or a database, but the virtualisation layer that runs all of them. This article looks back at what happened in summer 2026, explains why attackers go for vCenter and ESXi, and lists the controls that make the difference. For the wider context of the vulnerabilities exploited this year, read our analysis of the 2026 KEV entries.

Key Takeaways
  • Encrypting the datastores of one ESXi host stops every virtual machine on it at once: domain controllers, ERP, file servers and often the backup server too.
  • Broadcom fixed CVE-2026-59310 and CVE-2026-59309 (both CVSS 9.8) in vCenter on 29 July 2026; CISA confirmed exploitation of CVE-2026-59310, with known ransomware use, on 18 August.
  • CSIRT Italia's bulletin BL01/260528 describes Qilin affiliates using custom Rust ransomware built for ESXi, after entering through VPNs and perimeter devices.
  • The vCenter and ESXi management network must never be reachable from user VLANs, and the identities that administer it must not depend on the same Active Directory the attacker is trying to take over.
  • Backups the attacker can reach with stolen credentials are not backups. Qilin affiliates exploited Veeam CVE-2023-27532 to extract credentials and disable backups before encrypting.

Why attackers encrypt the hypervisor, not the virtual machines

A classic Windows ransomware attack has to spread across dozens or hundreds of machines: it needs administrative credentials on each one, it has to get past the endpoint protection installed on each one, and every host is a chance for the EDR to raise an alert. An attack on the hypervisor turns that problem upside down. A single ESXi host can run twenty, fifty or more virtual machines. If the attacker encrypts the virtual disk files on its datastores, every one of those machines stops at the same moment, without the ransomware ever running inside them.

There are three practical reasons why this has become the preferred route:

  • Impact per action. One compromised host, or one compromised vCenter that manages many hosts, takes down the whole data centre. For a company running everything on a three-node cluster, that means everything.
  • Less visibility. ESXi is not a general-purpose operating system and does not run the EDR agents you have on Windows servers. Many companies do not even forward its logs to a central system, so activity on the hypervisor goes unseen.
  • Recovery is harder. When the domain controllers, the backup server and the backup catalogue are themselves virtual machines on the encrypted datastores, the company has lost both production and the tools it needs to restore it.

Ransomware groups have understood this for years, and the KEV catalog reflects it. Entries flagged by CISA as known to be used in ransomware campaigns include CVE-2021-22005 in vCenter, CVE-2024-37085 in ESXi (added on 30 July 2024) and CVE-2025-22225 in ESXi (added on 4 March 2025). In January 2026 CISA also added CVE-2024-37079, an out-of-bounds write in vCenter's DCERPC implementation. 2026 added a new chapter.

The August 2026 vCenter case

On 29 July 2026 Broadcom published the advisory VMSA-2026-0006, covering ESX, vCenter, Workstation and Fusion. It was updated on 3 August with express patches for the 8.0 U2 line and on 19 August with guidance for 7.0. The advisory listed five vulnerabilities, with no workarounds for any of them:

CVEComponentTypeCVSS
CVE-2026-59309vCenter (VMware Directory Service)Authentication bypass9.8
CVE-2026-59310vCenter (Syslog server)Directory traversal9.8
CVE-2026-47876ESX (VMXNET3 adapter)Out-of-bounds write: a local administrator in a VM can execute code on the host9.3
CVE-2026-41703ESX, Workstation, FusionOut-of-bounds read7.6 (ESX)
CVE-2026-41709ESXInsufficient logging2.7

The fixed vCenter builds are 9.1.0.0300, 9.0.2.0100 and 8.0 U3k, with an 8.0 U2f express patch. For vCenter 7.0, the advisory tells customers with an extended support contract to contact Broadcom Support. When it was published, the advisory did not mention exploitation in the wild.

That changed three weeks later. On 18 August 2026 CISA added CVE-2026-59310 to the KEV catalog, describing it as a path traversal that "could allow a threat actor with network access to vCenter to execute arbitrary code", and set its known ransomware campaign use to Known. On 28 October CISA also added CVE-2026-41709, the ESX logging flaw that lets an administrator perform certain operations without leaving a trace. On its own it scores only 2.7, but in the hands of an attacker who already has admin access it helps cover tracks.

In Italy, CSIRT Italia published an alert on CVE-2026-59309 and CVE-2026-59310, and the ACN Operational Summary for August 2026 records both the alert and the KEV addition. The same summary notes that ransomware incidents rose in August, that their confirmed impacts included the unavailability of data and systems and, in some cases, the involvement of backup infrastructure, and that the main entry vectors were known vulnerabilities and valid credentials stolen earlier.

The key detail is in CISA's wording: network access to vCenter. vCenter is rarely exposed directly to the internet. The scenario that matters is an attacker who is already inside, through a VPN, a perimeter device or a phished workstation, and who can reach the vCenter interface from the network they landed on. That is a network design problem before it is a patching problem.

What CSIRT Italia saw: Qilin and ESXi

On 28 May 2026 CSIRT Italia published bulletin BL01/260528 on the Qilin group. It reports a significant number of Qilin attacks in Italy since the start of 2026, mostly against SMEs but including important cloud service providers. The bulletin reads like a description of the full chain that ends on the hypervisor:

  • Initial access through perimeter devices exposed to the internet and left unpatched: Ivanti EPMM (CVE-2026-1281 and CVE-2026-1340) and Fortinet (CVE-2024-21762 and CVE-2024-55591), plus brute force against SSL VPNs such as SonicWall and access bought from initial access brokers. CSIRT Italia names the absence of multi-factor authentication as one of the main structural weaknesses of the victims.
  • Suppliers as a way in, through compromised cloud services and VPN credentials.
  • Selective encryption of hypervisors and backups: custom ransomware written in Rust, designed for VMware ESXi, with PowerShell scripts that automate the encryption of vSphere datastores. Before that, the attackers exploited Veeam Backup & Replication CVE-2023-27532 to extract credentials and neutralise online backups.
  • Anti-forensics: legitimate remote administration tools (Atera, Splashtop, ScreenConnect) as command and control, systematic deletion of logs, disabling of EDR and antivirus, and exfiltration with Rclone, WinSCP and FileZilla.

If you want to follow one of these chains from start to finish, we reconstructed a VPN-to-ransomware case in From VPN to ransomware: the zero-day exploited by a Qilin affiliate. The point for this article is simple: the hypervisor is the last step, and most of the defensive work happens in the steps before it.

Keep the management network isolated

The single most effective control against hypervisor ransomware is architectural: vCenter, the ESXi management interfaces and the out-of-band controllers (iDRAC, iLO, IPMI) must live in a dedicated management network that user VLANs cannot reach.

What that means in practice

  • A separate VLAN or VRF for management, with a firewall in front of it, not just an access list on the core switch.
  • Access only from dedicated administration workstations or a jump host, with MFA, never from ordinary office PCs and never directly from the VPN address pool.
  • No routing from the server VLANs, Wi-Fi or guest networks towards management ports (443 and 902 on ESXi, 443 and 5480 on vCenter, the web interfaces of out-of-band controllers).
  • Outbound internet traffic from vCenter and ESXi blocked or limited to what updates really need. An ESXi host has no reason to talk to a file-sharing service.

With that segmentation in place, a flaw like CVE-2026-59310 does not disappear, but the attacker who lands on a phished laptop can no longer reach the vulnerable service. That buys the time you need to patch.

Hardening the hosts themselves

  • Lockdown mode on ESXi, so the host is managed only through vCenter or a short list of exception users.
  • SSH and ESXi Shell disabled by default and turned on only for a maintenance window, with an alert whenever they are enabled.
  • The execInstalledOnly setting, which prevents the host from running binaries that were not installed through a signed package. It is designed to block exactly the kind of custom encryptor that lands on the host and runs from a datastore.
  • Patching vCenter and ESXi on a cycle measured in days for KEV entries, not in quarters.

Active Directory and vCenter SSO: the identity path to the hypervisor

Most hypervisor attacks do not need a hypervisor exploit at all. The attacker takes over Active Directory, then uses legitimate credentials to log into vCenter or into the hosts. If vCenter SSO trusts your AD and the "VMware administrators" group is an AD group, then a domain admin is a vCenter admin.

CVE-2024-37085 showed how dangerous that coupling can be: on ESXi hosts joined to Active Directory, a malicious actor with sufficient AD permissions could gain full administrative access to the host. CISA lists it as used in ransomware campaigns. The lesson goes beyond that single CVE:

  • Do not join ESXi hosts to Active Directory unless you have a concrete reason. If they are joined, check how the host treats the "ESX Admins" group and who can create or modify it.
  • Use dedicated administrative accounts for vCenter, separate from the accounts used for email and daily work, and protect them with MFA through vCenter's identity federation.
  • Treat vCenter as Tier 0, like domain controllers: whoever can administer it can stop the company.
  • Close the paths to Tier 0 in AD: Kerberoastable service accounts with weak passwords, unconstrained delegation, excessive ACLs, accounts with DCSync rights that should not have them, an old krbtgt password. These are the steps an attacker climbs before reaching vCenter.

Backups the attacker cannot reach

The Qilin bulletin describes a deliberate step: before encrypting the datastores, the attackers exploited Veeam CVE-2023-27532 to extract credentials and neutralise the online backups. The ACN August summary also mentions incidents in which the backup infrastructure was involved. A backup that the attacker can delete with the credentials they have just stolen does not protect you.

ControlWhy it matters against hypervisor ransomware
At least one immutable or offline copyObject-lock storage, a hardened repository or media that is physically disconnected cannot be wiped from the console.
Backup server outside the production domainIf it is joined to the same AD, a domain admin is also a backup admin.
Backup server not running on the cluster it protectsOtherwise the backup is encrypted along with the virtual machines it was meant to restore.
Patched backup softwareCVE-2023-27532 had been fixed for years when it was used in 2026.
MFA and separate credentials for the backup consoleStolen AD credentials should not open it.
Restore tests that include domain controllers and vCenterRestoring a file proves little. Restoring the whole environment from zero is what you will have to do.

Detection signals: what to watch on vCenter and ESXi

Hypervisor attacks move fast once the attacker has admin access, which is why the time between the first intrusion and the encryption is what decides the outcome. We covered why dwell time matters so much in Mean Time to Detect: why dwell time is your biggest security risk. On the virtualisation layer, these are the signals worth an immediate alert:

SignalWhat it may mean
SSH or ESXi Shell enabled outside a maintenance windowSomeone is preparing to run commands directly on the host.
Use of esxcli or vim-cmd to list or power off VMs in bulkVirtual machines are being stopped so their disk files can be encrypted.
Mass power-off of virtual machinesOften the last step before encryption.
New local accounts on ESXi or new vCenter administratorsPersistence, or preparation of an access that survives a password reset in AD.
Change to execInstalledOnly or lockdown modeProtections being disabled so an unsigned binary can run.
Mass renaming of files in the datastoresEncryption in progress.
Log forwarding from ESXi or vCenter stopsSomeone is cleaning up, as the Qilin affiliates did.
Remote administration tools appearing on servers (Atera, Splashtop, ScreenConnect)A command and control channel that looks legitimate.

None of these signals is visible if vCenter and ESXi logs stay on the hosts. Forward them by syslog to a SIEM or SOC that is watching 24 hours a day, and make sure someone is alerted when the forwarding itself stops.

Checklist: ten steps for your virtual infrastructure

  1. Verify that every vCenter is on a fixed build (9.1.0.0300, 9.0.2.0100, 8.0 U3k or the 8.0 U2f express patch), and plan the exit from 7.0 if you still run it.
  2. If vCenter was unpatched and reachable from user networks between late July and the update, treat it as a possible compromise: review administrators, local accounts and logs, and rotate the vCenter and ESXi credentials.
  3. Move vCenter, ESXi management and out-of-band controllers into an isolated management network, reachable only from administration workstations with MFA.
  4. Enable lockdown mode and execInstalledOnly, and keep SSH and ESXi Shell disabled.
  5. Decide whether ESXi really needs to be joined to AD. If it does, review the "ESX Admins" group and who controls it.
  6. Use dedicated vCenter admin accounts with MFA, and treat vCenter as Tier 0.
  7. Assess Active Directory for the paths that lead to Tier 0: Kerberoasting, delegation, ACLs, DCSync rights, krbtgt age.
  8. Keep at least one immutable or offline backup copy, with the backup server outside the production domain and outside the cluster it protects.
  9. Forward ESXi and vCenter logs to a monitored SIEM and alert on the signals in the table above.
  10. Test a full restore, including domain controllers and vCenter, and write the steps into your incident response plan.

How Orizon can help

Most of the steps above start with knowing what is actually on your internal network. RECON Internal uses a Scout probe installed on a Linux host inside your network, which connects out over HTTPS only. It finds every device in the subnets you choose, including servers and management interfaces, reads installed software and missing patches, and re-tests each finding before reporting it as Proven, To verify or Closed. Its Active Directory assessment checks Kerberos (Kerberoasting, AS-REP roasting, krbtgt password age), delegation, ACLs and DCSync rights, Tier-0 groups and AD CS templates, and ranks the attack paths to Tier 0. It detects these conditions; it never runs the exploits.

Detection on the virtualisation layer needs people watching. Oversight is a 24/7 SOC on the CYREBRO platform: analysts validate every detection, and Orizon installs the service on your stack, monitors it with you and intervenes when an attack needs a response, with containment, credential resets, hardening and backup checks as agreed. If you suspect the hypervisor has already been touched, start from your incident response plan and from the exploited vulnerabilities listed in our 2026 KEV analysis.

ransomware
esxi
vcenter
vmware
hypervisor
backup
active-directory
kev