Signs of Attacks on Backup Infrastructure and Alerts

Which operations on backup infrastructure should raise alerts? Tying job deletion, retention shortening, shadow copy deletion and repository changes to SIEM alerts.
Technical summary
Technology
SIEM, Sysmon, Windows event logs, VSS, VMware ESXi, S3 Object Lock, CloudTrail, Sigma, MITRE ATT&CK T1490
Topic
Early detection of attacks on backup infrastructure
Level
Intermediate–advanced
Prerequisites
  • Central collection of backup, hypervisor and storage logs
  • Process creation logging with command lines (4688 or Sysmon)
  • An on-call and responder defined for alerts
In this article7 sections

Before running the encryption command, a ransomware operator closes off your recovery paths: deletes backup jobs, shortens retention periods, destroys shadow copies and recovery settings, and opens access to hypervisors. This preparation is often done hours, even days, before encryption. If you’re monitoring the right events, you’ll see it. This guide turns these behaviors into detectable events, alert priorities and response steps.

Why attackers target backups first

The bargaining power of a ransom demand rests on the organization not being able to recover on its own. A working backup neutralizes the attacker’s only trump card. That’s why backup infrastructure is the target that comes right after credential theft and lateral movement in the attack chain. Backups first, then encryption. MITRE ATT&CK groups these behaviors under technique T1490 “Inhibit System Recovery”.

The typical sequence seen in the field:

  1. Domain admin rights or backup console credentials are obtained.
  2. A session is opened on the backup console; repositories, jobs and copy targets are discovered.
  3. Jobs are disabled or deleted, retention periods are shortened, restore points are deleted.
  4. On Windows systems, shadow copies and the recovery environment are disabled.
  5. Hypervisors are accessed over SSH or the management interface; virtual machines are shut down and encrypted.

Every event between the second and fourth steps is a detection opportunity; if this window is missed, the second chance is usually an encrypted screen. Preventive design (separating backup infrastructure in terms of identity) is the subject of the identity isolation in backup infrastructure guide. The goal here is to see the attack early not where prevention holds, but where it doesn’t.

Backup console events to monitor

Most backup software writes administrative operations to its own audit log, the Windows event log or syslog. Event IDs and field names vary by product and version; don’t trust an event list copied from another environment. First build the event catalog for your own product, then map the records corresponding to the behaviors below:

  • Session events: Successful and failed console sign-ins, unusual source addresses, after-hours sessions.
  • Job changes: Job deletion, disabling, removing the schedule, removing virtual machines from scope.
  • Retention changes: Reducing the number of restore points or days, removing GFS settings.
  • Repository changes: Removing a repository, shortening the immutability period, deleting a copy target.
  • Restore point deletion: The delete from disk command, mass deletion of backup files.
  • Security settings: Changing the encryption password, adding users or roles, turning off MFA, changing the four-eyes approval setting.
  • Notification settings: Changing the email or syslog destination; the attacker may try to turn the volume down first.

These events are only useful when sent outside the console in real time. An attacker who takes over the console deletes local records too. After setting up forwarding to the SIEM via syslog or an agent, make the interruption of forwarding a separate alarm too. A source that goes quiet is a suspicious source.

Shadow copy and VSS indicators on the Windows side

The commands used to inhibit recovery on Windows endpoints and servers are well known; if command-line logging is on, they’re easy to catch. The prerequisite is that the policy to record the command line in process creation events (Security log 4688) is enabled, or that Sysmon event ID 1 is collected. A common gap: 4688 is being collected, but the command-line field is empty.

Command / behavior Purpose Source
vssadmin delete shadows /all /quiet Deleting shadow copies 4688, Sysmon 1
vssadmin resize shadowstorage (very small value) Triggering automatic deletion of shadow copies 4688, Sysmon 1
wmic shadowcopy delete or deleting Win32_ShadowCopy with PowerShell Deleting shadow copies 4688, Sysmon 1, PowerShell 4104
wbadmin delete catalog -quiet Deleting the Windows Server Backup catalog 4688, Sysmon 1
bcdedit /set {default} recoveryenabled no Turning off the Windows recovery environment 4688, Sysmon 1
Stopping or disabling the VSS service Preventing backups from being taken System 7036 / 7040
Clearing the Security or System log Covering tracks Security 1102, System 104

The Sigma rule below looks for shadow copy deletion attempts in process creation events. Sigma is a SIEM-independent rule format and can be converted into your SIEM’s query language:

title: Shadow Copy Deletion Attempt
status: experimental
logsource:
  category: process_creation
  product: windows
detection:
  sel_vssadmin:
    Image|endswith: '\vssadmin.exe'
    CommandLine|contains|all:
      - 'delete'
      - 'shadows'
  sel_wmic:
    Image|endswith: '\wmic.exe'
    CommandLine|contains|all:
      - 'shadowcopy'
      - 'delete'
  sel_resize:
    Image|endswith: '\vssadmin.exe'
    CommandLine|contains|all:
      - 'resize'
      - 'shadowstorage'
  condition: 1 of sel_*
level: high
tags:
  - attack.impact
  - attack.t1490

Legitimate use is rare but exists: some backup products and disk maintenance scripts manage shadow copies. Before putting the rule live, run it on a few weeks of historical data and add legitimate parent processes and service accounts as exceptions. Define exceptions not by file name alone but together with the parent process path and account; an attacker can bring a tool with the same file name. A broad exception means a blind spot.

Indicators on the storage and hypervisor side

Hypervisor

Ransomware targeting VMware ESXi usually gets into the ESXi shell over SSH, shuts down virtual machines and encrypts disk files. One host means dozens of servers. What to monitor:

  • SSH or the ESXi Shell service being enabled on ESXi (visible in vCenter events and host logs).
  • Lockdown mode being turned off and new local users being created.
  • Commands in ESXi shell.log and auth.log that shut down many virtual machines or kill processes within a short time.
  • Mass power-offs, snapshot deletions and datastore file operations in vCenter.

Send these logs from ESXi to a remote syslog destination. If ESXi’s local logs aren’t on persistent storage, they’re lost on restart; hosts left without the scratch and log location being checked at installation are often found in the field.

Storage and object storage

On NAS and storage arrays, sign-ins to the management interface, snapshot deletion, removing snapshot schedules and stopping replication are monitored. In S3-compatible object storage, configuration changes leave clear traces. Example events to monitor through CloudTrail in AWS:

  • PutBucketLifecycle: Adding a lifecycle rule that will automatically delete objects (although the API name is PutBucketLifecycleConfiguration, the CloudTrail event is recorded under this name).
  • PutBucketVersioning: Suspending versioning.
  • PutObjectLockConfiguration: Changing the bucket-level default retention setting (a management event).
  • PutObjectRetention: Changing the retention period at object level; in Governance mode, attempts to shorten it with the s3:BypassGovernanceRetention permission appear in this event. Because it’s object-level, it’s a data event.
  • DeleteObject and DeleteObjects: Mass deletion. This and the object-level events above are data events in CloudTrail; they aren’t logged by default and must be enabled separately for the backup bucket.

Which attacker privilege the immutable repository options protect against is compared in the immutable backup options article. Monitoring makes visible the moment someone tries to get around these protections. Protection and monitoring work together.

An event–alert priority table

Sending every event at the same priority leads the on-call person into alert fatigue, and the real alert gets lost in the noise. The table below suggests a starting priority based on how strong an attack indicator the event is on its own and whether its effect can be undone:

Event Priority Reason
Shortening the immutability period, removing a repository Critical Targets the protection itself
Deleting a backup job or restore point Critical Irreversible data loss
Changing the encryption password Critical Can tie new backups to the attacker’s key
Deleting shadow copies, turning off the recovery environment High Seen right before encryption
Enabling ESXi SSH, turning off lockdown High A preliminary step to hypervisor encryption
Shortening retention, disabling a job High Quiet but cumulative effect
A console sign-in from a new source address or after hours Intermediate Weak on its own, strong in correlation
A change of notification or syslog destination, interruption of the log flow Medium–high Blinding the monitoring

In enterprise cybersecurity monitoring, correlation produces a strong signal from individual events. An after-hours console sign-in followed within an hour by a retention change with the same account: two medium-priority events should become a single critical alert. Weak alone, clear together.

Sending to the SIEM and response actions

There are three rules in forwarding design. First, the log flow is independent of the backup infrastructure’s identity domain; administrator access to the SIEM isn’t the same as the backup administrator accounts. Second, the “last event time” is monitored per source, and an alarm is raised for a source that doesn’t send logs within the expected time. Third, the retention period for backup events in the SIEM is kept long enough to cover a possible investigation; this topic is covered in the central log retention architecture article.

Response steps for critical and high-priority alerts are written in advance; decisions aren’t made by thinking it through at three in the morning:

  1. Identify the account that made the change and the session source; confirm against the change calendar and its owner.
  2. If it can’t be confirmed, disable the account and restrict access to the backup console and hypervisor management to the management network.
  3. Keep the event records and the current state of the relevant configuration as evidence; how to do this is explained in the preserving forensic evidence after a cyber incident guide.
  4. Return retention, immutability and encryption settings to a known-good state and recreate deleted jobs.
  5. Start the incident response process to look for traces of lateral movement and privilege escalation across the domain.

The work doesn’t end once the rules go live. In a test environment, an authorized person deliberately generates these events, and it’s regularly verified that the alert reaches the SIEM and is passed to the on-call person. An untested alert is an assumption. If you want to design detection rules and the log flow for your backup infrastructure together, see the scope of our SIEM deployment and security logs and backup security and ransomware readiness services.

Technical reference: MITRE ATT&CK T1490: Inhibit System Recovery.

Frequently asked questions

Are the backup software’s own email notifications enough for monitoring?

No. These notifications are mostly built around job success and failure; they don’t cover administrative events such as job deletion, retention period changes or encryption setting changes. An attacker who takes over the console can also turn off notification settings. Send events in real time to a SIEM outside the console.

Is a shadow copy deletion event always a sign of attack?

No. Some backup products and maintenance scripts manage shadow copies legitimately. That’s why the rule should focus on calls other than known backup processes and service accounts; before writing the rule, extract the legitimate uses from a few weeks of historical data and add them to an exception list.

If we use an immutable repository, do we still need this monitoring?

Yes. Immutability only protects existing restore points. An attacker can stop new jobs, change the encryption password, turn off copy jobs or shorten the immutability period for subsequent writes. If these events go unnoticed, within a few weeks there may be no current, usable backup left.

What should be done first on a critical alert from the backup console?

The on-call person first verifies whether the account that made the change and the session source are legitimate. If this can’t be verified, the account is disabled, access to the console is restricted to the management network, and the event record and a screenshot are kept as evidence before the change is reverted. Then the incident response process is started.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Cybersecurity

Firewall, VPN, endpoint security and threat prevention. We reduce the attack surface and measure and report the risks.

Knowledge Center

Other articles