Join our Newsletter — 33% off our NHI Course
Home› Guides› ESXi Logs Detection Guide: hostd, vmkernel, Shell and…
Guide SOC & Detection Engineering

ESXi Logs Detection Guide: hostd, vmkernel, Shell and Audit Records for the SOC

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 8 November 2026 14 min read
On this page

Ransomware crews and espionage actors target VMware ESXi because one hypervisor holds dozens of servers and no EDR agent runs on it, so its own logs are the only record of what an intruder did there. By default ESXi keeps those logs in memory, loses them on reboot, holds roughly a day of history, and does not produce audit records at all. This guide explains which ESXi log files matter (hostd, vmkernel, shell, auth, vpxa, esxupdate and the audit record stream), how to forward them securely, which events reveal the attack chain from SSH enablement to mass VM power-off, and how vCenter events complete the picture. It is written for detection engineers, SOC analysts and virtualisation administrators. Host settings are covered in the VMware ESXi Hardening Guide; this page is about the logs.

This is an independent guide from NHI Mgmt Group, based on Broadcom's public documentation as of November 2026. It is not sponsored or reviewed by Broadcom. Product features change, so confirm settings against Broadcom's current documentation.

Key takeaways

  • Persist and forward first. ESXi logs live in an in-memory file system by default and are lost on reboot; set Syslog.global.logHost to a TLS collector and a persistent Syslog.global.logDir before anything else.
  • Turn audit records on. They are disabled by default; Syslog.global.auditRecord.storageEnable and remoteEnable add source IPs and command outcomes that ordinary logs lack.
  • auth.log plus shell.log tells you who ran what. Join them on the session's world ID to attribute every ESXi Shell command to a login and source IP.
  • The attack chain is short and loud. SSH service start, a firewall or syslog change, execInstalledOnly disabled, file transfer to /tmp, then many VM power-off operations: each step is a detection.
  • ESXi logs are not enough alone. Correlate with vCenter events, Active Directory group changes and help desk tickets, because most intrusions reach ESXi through identity, not exploits.

The ESXi log files and what each one records

All current ESXi releases write logs under /var/run/log (with /var/log links). Broadcom's knowledge base lists the files below; the ATT&CK v20 log source names show how MITRE refers to them in its analytics.

FileATT&CK v20 log sourceContentsDetection value
hostd.logesxi:hostdHost management agent: VM and host tasks and events, vSphere Client, vpxa and SDK sessionsVM power operations, reconfiguration, service starts, API logins
vmkernel.logesxi:vmkernelVMkernel: devices, storage and network drivers, VM start-upKernel setting changes, module loads, datastore activity
shell.logesxi:shellESXi Shell enabled and disabled, and every command typed at the console or over SSHHands-on-keyboard commands such as esxcli and vim-cmd
auth.logesxi:authShell and SSH authentication successes and failures, source IP, method (password or public key), session open and closeWho logged in, from where, and with which method
vobd.logesxi:vobdVMkernel Observation eventsHost-level alarms and state changes
syslog.logesxi:syslogManagement service start-up, watchdogs, scheduled tasks, DCUI useLocal console use, service restarts
vpxa.logesxi:vpxavCenter agent communication with vCenter and hostdActions pushed from vCenter, account operations
esxupdate.logesxi:esxupdatePatch and VIB installationForced or unsigned VIB installs
vmkwarning.log(vmkernel)Warnings and alerts from the VMkernel logClock changes around attacker activity (timestomping)
vmksummary.log(vmkernel)Start-up and shutdown summary and an hourly heartbeatUnexpected reboots; gaps in the heartbeat
Audit records(audit)Security-relevant actions with outcome, written to /scratch/auditLog by default and optionally forwardedCommand result (for example exit status 126 on a blocked binary), login source
Per-VM vmware.logn/aVM process log in the VM's datastore folder, including guest operationsHost-to-guest command execution and file transfer (GuestOps entries)

Two other artefacts matter in investigations: /var/db/esximg/profile, which records every installed VIB with time and install flags, and /var/log/boot.gz, the compressed boot log.

Forwarding and persisting ESXi logs

The default is fragile: logs sit in an in-memory file system, survive no reboot and hold about 24 hours. Advanced syslog options are available from ESXi 7.0 Update 1 and are set as host advanced settings (vSphere Client, PowerCLI Get-AdvancedSetting or esxcli system syslog config set).

SettingRecommended valueNotes
Syslog.global.logHostssl://collector.example:1514 (two collectors, comma-separated)Format protocol://host[:port]; defaults 1514 for SSL and TCP, 514 for UDP; keep to five hosts or fewer; blank means nothing is forwarded
Syslog.global.certificate.checkSSLCertstrueReplaces the deprecated logCheckSSLCerts; requires the collector's CA certificate on the host; optionally enable CRL checking and strict X.509
Firewall ruleset syslogEnabled, restricted to collector addressesesxcli network firewall ruleset set --ruleset-id=syslog --enabled=true, then refresh
Syslog.global.logDirPersistent location (local /scratch or a datastore)Only /scratch on local storage persists across reboots; with shared storage set Syslog.global.logDirUnique to true
Syslog.global.auditRecord.storageEnabletrueLocal audit records in Syslog.global.auditRecord.storageDirectory (default /scratch/auditLog)
Syslog.global.auditRecord.storageCapacitySized for your retention (MiB)Can be increased but not decreased
Syslog.global.auditRecord.remoteEnabletrueSends audit records to the logHost destinations; required by the DISA STIG for ESXi 8.0 (ESXI-80-000233)
  1. Pilot: apply the settings to one cluster through host profiles or configuration management, run esxcli system syslog reload, and confirm every file type arrives at the collector with correct host names and timestamps.
  2. Test loss: reboot a pilot host and confirm pre-reboot logs exist on the collector and in the persistent directory.
  3. Roll out by cluster and add a compliance check that reports any host whose logHost or audit settings drift. Rolling back is a single advanced-setting change, but treat any unplanned change of these values as a security event.
  4. Collect vCenter too: forward vCenter Server Appliance syslog and vCenter events. Mandiant notes the appliance does not forward denied connections or shell activity by default.
  5. Retain long: keep at least a year in the SIEM; hypervisor intrusions are often found late.

Detections across the attack chain

Mandiant's July 2025 analysis of UNC3944 (Scattered Spider) vSphere intrusions and its earlier work on malicious VIBs describe a consistent sequence. The table maps each step to the logs that show it.

StageWhat to look forWhereATT&CK v20
Access through identityNew member of an AD group that grants vSphere admin (for example "ESX Admins"); vCenter login success from an unusual sourceAD event 4728 or 4732; vCenter UserLoginSessionLoginSuccessEventT1078.002 Domain Accounts
SSH enabledSSH or ESXi Shell service started on a hosthostd.log, shell.log; vCenter HostServiceStartedEventT1021.004 SSH
Root loginRoot SSH login, especially password authentication from a non-bastion IPauth.log, audit recordsT1078 family; T1059.012 Hypervisor CLI
Root password or account changesNew local accounts or role changesshell.log (esxcli system account add), vpxa.log, hostd.logT1136.001 Local Account, T1098 Account Manipulation
Defence impairmentesxcli system syslog config set or reload; firewall changes; execInstalledOnly set to FALSEshell.log, audit records, vCenter HostFirewallInfoChangedEventT1685 Disable or Modify Tools, T1686 Disable or Modify System Firewall
Tool transferSCP or SFTP of binaries to /tmp or a datastore; chmod on new filesauth.log, shell.log, flow logsT1570 Lateral Tool Transfer
PersistenceVIB installed with --force or a lower acceptance level; edits to /etc/rc.local.d/local.sh or cronesxupdate.log, VIB profile, shell.logT1505.006 vSphere Installation Bundles, T1037.004 RC Scripts, T1053.003 Cron
Credential theft by diskDomain controller powered off, its VMDK attached to another (often orphaned) VM, then powered onvCenter VmPoweredOffEvent, VmReconfiguredEvent, VmPoweredOnEvent; hostd.logT1006 Direct Volume Access, T1003.003 NTDS
Backup sabotageSnapshot and backup deletion; backup admin group changeshostd.log, backup software logs, ADT1490 Inhibit System Recovery
ImpactMany vim-cmd vmsvc/power.off commands in minutes; files renamed with new extensions; long nohup jobsshell.log, hostd.log, vmkernel.logT1489 Service Stop, T1486 Data Encrypted for Impact, T1529 System Shutdown/Reboot
Anti-forensicsClock changes, cleared or rotated logs, syslog daemon stopped, history removedvmkwarning.log, gaps at the collectorT1070.006 Timestomp, T1070.003 Clear Command History, T1690 Prevent Command History Logging

Rules worth building first

  • SSH start without a ticket: any SSH or ESXi Shell service start, joined to change records. In a well-run estate SSH is off and starts are rare, so this alert is high-fidelity. With lockdown mode enabled, blocked SSH attempts are themselves a strong signal.
  • Root login from outside the bastion: parse auth.log for source IP and method; alert on any root login not from an approved jump host and on password logins where keys are standard.
  • Attributed command stream: link auth.log sessions to shell.log commands through the world ID, as Broadcom documents, and alert on high-risk commands: esxcli system syslog, esxcli network firewall, esxcli system settings kernel set -s execInstalledOnly, esxcli software vib install with --force or --no-sig-check, esxcli system account and vim-cmd vmsvc/power.off.
  • Mass power-off: more than a small number of VM power-off operations from one session or host in ten minutes, outside a maintenance window. Page on-call; this is often minutes before encryption.
  • Critical VM disk moves: a reconfiguration event that attaches a domain controller or other Tier 0 VMDK to a different VM, or any power-on of VMs in decommissioned or orphaned folders.
  • Logging change or silence: any change to Syslog.global.logHost or the audit record settings, and any host that stops sending for longer than its baseline.

Response steps once these fire are in the ESXi Ransomware Playbook.

Where identity fits

Most hypervisor compromises begin with an identity: a reset help desk account, an AD group that maps to vSphere admin, or a vCenter service account. ESXi logs show the root or vpxuser session, not the person behind it, so the investigation has to work backwards through vCenter and the directory.

  • AD-joined hosts: any change to the AD group configured as the ESXi admin group should alert at the same priority as an ESXi root login. Mandiant recommends considering removal of ESXi and vCenter from AD altogether.
  • vpxuser: vCenter manages hosts through the vpxuser account. On ESXi 8.0 and later Mandiant recommends esxcli system account set -i vpxuser -s false so that a compromised vCenter user cannot change ESXi root passwords.
  • Root and break-glass credentials: vault them, rotate after use, and alert on any root login that is not paired with a vault checkout. See the Privileged Access Management Guide.
  • Automation and backup service accounts: backup proxies and monitoring tools log in to hosts constantly; baseline their source IPs and API calls so a stolen service credential stands out. See the Help Desk Social Engineering Incident Playbook for the front of this chain.

Measuring coverage and choosing an approach

Decision guidance: a small estate with a handful of standalone hosts can forward directly from each host to the SIEM over TLS and rely on simple rules (SSH start, root login, power-off bursts). A large estate should manage syslog and audit settings centrally through host profiles or configuration management, collect vCenter events as a separate structured feed, and build correlation across AD, vCenter and ESXi. Use TLS (ssl:// on 1514) wherever possible; TCP 514 without TLS is acceptable only on an isolated management network; avoid UDP because messages can be dropped silently during exactly the bursts you need to see.

  • Percentage of hosts forwarding all log types over TLS to two collectors (target 100 per cent).
  • Percentage of hosts with audit records enabled locally and remotely (target 100 per cent).
  • Hours per month SSH is running across the estate, and the share of SSH starts matched to a ticket.
  • Time from a test SSH start or test power-off burst on a pilot host to SOC alert (target under 15 minutes).
  • Hosts with a logging gap longer than one hour in the last 30 days (target zero unexplained).
FrameworkRelevant controls
NIST CSF 2.0PR.PS-04 log records generated and available, DE.CM-09 monitoring of hardware, software and runtime environments, DE.AE-02 analysis, DE.AE-03 correlation
NIST SP 800-53 Rev. 5AU-2, AU-3, AU-4 Audit Log Storage Capacity, AU-6, AU-9, AU-11, AU-12, SI-4, CM-3
CIS Controls v8.1Control 8 Audit Log Management
ISO/IEC 27001:2022 Annex A8.15 Logging, 8.16 Monitoring activities, 8.17 Clock synchronisation
DISA STIGVMware vSphere 8.0 ESXi STIG, including ESXI-80-000233 (off-load audit records via syslog)

Common failure patterns

  • Logs only in memory: the host is rebooted during response or by the attacker, and the only copy of shell and auth history disappears.
  • Audit records never enabled: they are off by default, so command outcomes and login sources are missing when the investigation needs them.
  • vCenter collected, hosts not: attackers who enable SSH and work directly on hosts bypass vCenter's event stream; only host logs show their commands.
  • UDP syslog under load: a mass power-off produces bursts of messages, and UDP drops some of them without trace.
  • No identity correlation: the SOC sees a root session but cannot tie it to the AD group change and help desk reset that made it possible, so containment misses the real entry point.
  • Logging changes not alerted: attackers change logHost or stop the syslog daemon early, and the resulting silence is treated as a collector issue.

Lessons from incidents

  • UNC3944 vSphere campaigns (2025): help desk resets led to an AD group grant for vSphere admin, vCenter appliance takeover, SSH on hosts, offline theft of a domain controller's NTDS.dit by moving its disk, backup deletion and encryption. Every hypervisor step is visible in the logs on this page, but only if they were forwarded and correlated with directory events. See the Scattered Spider Threat Actor Profile.
  • Malicious VIB persistence (2022): Mandiant found backdoors installed as VIBs with --force, a stopped syslog daemon and altered clocks. esxupdate.log, the VIB profile and vmkwarning.log held the evidence; forwarding them off-host removes the attacker's ability to clean up.
  • Ransomware groups targeting ESXi: groups profiled in the Akira Threat Actor Profile and the Ransomware Groups Guide target ESXi directly; the power-off and encryption rules above are the last chance to interrupt them.

Practitioner checklist

  • Set Syslog.global.logHost to two TLS collectors and enable certificate checking.
  • Set a persistent Syslog.global.logDir (with logDirUnique on shared storage).
  • Enable audit records locally and remotely and size their storage for your retention target.
  • Forward hostd, vmkernel, shell, auth, vpxa, syslog, esxupdate and vmkwarning logs, plus vCenter events and appliance syslog.
  • Attribute shell commands to logins by joining auth.log and shell.log on world ID.
  • Alert on SSH starts without a ticket, root logins outside the bastion and changes to syslog, firewall or execInstalledOnly.
  • Alert on VM power-off bursts and on Tier 0 disks attached to other VMs.
  • Correlate ESXi and vCenter activity with AD group changes and help desk password resets.
  • Monitor a logging heartbeat per host and treat silence as a security event.
  • Test each rule quarterly on a pilot host and record time to alert.

Standards and references

Related NHI Mgmt Group resources: VMware ESXi Hardening Guide · VMware vCenter Server Hardening Guide · ESXi Ransomware Playbook · Hypervisor and Virtualisation Security Guide · Detection Engineering Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 8 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.