Before ransomware encrypts or a wiper destroys, the attacker almost always removes the ways back: shadow copies, backup catalogues, recovery settings, snapshots and cloud object versions. MITRE ATT&CK tracks this as T1490 Inhibit System Recovery in the Impact tactic (TA0040), across Windows, Linux, macOS, ESXi, network devices, containers and IaaS. Because legitimate deletion of every shadow copy on a host is rare, this step is one of the most reliable late-stage tripwires defenders have, and protecting recovery data makes it fail outright. This guide describes the technique at the level of the ATT&CK description, then concentrates on detection, protection of recovery points and response. It is for detection engineers, backup and infrastructure teams and incident responders. Broader ransomware controls are in the Ransomware Prevention Guide.
Key takeaways
- It is a high-fidelity signal. Native utilities deleting all shadow copies, removing the backup catalogue or disabling Windows recovery, especially across several hosts at once, should trigger automated containment, not a ticket.
- The cloud and hypervisor versions matter as much. Deleting snapshots, disabling versioning, shortening backup retention and scheduling lifecycle deletion are the same technique in a control plane.
- Immutability defeats it. Recovery points that no identity can delete before expiry, including administrators and the root account, turn T1490 from catastrophic to noisy.
- Backup identities are the target. Attackers reach backups through administrator accounts, backup service accounts and cloud keys; separate them from the production identity plane.
- Shadow copies are not backups. They live on the same volume and are deleted in seconds; plan recovery around offline or immutable copies that have been restore-tested.
How the attack works
ATT&CK describes T1490 as adversaries deleting or removing built-in data and turning off services designed to help recover a corrupted system, to deny access to backups and recovery options. It is used to amplify Data Destruction (T1485) and Data Encrypted for Impact (T1486), and attackers may also disable recovery notifications before corrupting backups. The ATT&CK v20 description groups the behaviour by platform:
| Platform | What is removed or disabled | Native mechanisms ATT&CK names |
|---|---|---|
| Windows | Volume shadow copies, Windows Backup catalogue, automatic recovery, Windows Recovery Environment | vssadmin.exe, WMI (wmic shadowcopy), wbadmin.exe, bcdedit.exe, REAgentC.exe, diskshadow.exe |
| Linux | Rescue targets, fstab backups, EFI partition content, snapshot recovery | Shell utilities and scripts; chattr per the MITRE Linux analytic |
| ESXi | Virtual machine snapshots | vim-cmd snapshot removal or deletion of snapshot files |
| Network devices | Backup firmware images and file systems | Disk Wipe (T1561) then System Shutdown/Reboot (T1529) |
| IaaS and storage | Versioning, backup policies, snapshots, database backups, machine images, prior object versions | Cloud APIs and CLIs using compromised credentials |
| Any | "Online" backups on network shares or synced cloud folders | Ordinary file deletion or encryption |
The Windows variants need administrative rights on the host, which is why T1490 normally appears late, after privilege escalation and lateral movement, and often pushed to many hosts at once by group policy, remote execution or a deployment tool.
Where it shows up
- Ransomware families: ATT&CK records shadow copy deletion by Ryuk (S0446), BlackCat (S1068), Clop (S0611), Black Basta (S1070) and LockBit 2.0 (S1199), and records BlackCat and Clop also changing boot configuration to disable recovery. Wizard Spider (G0102) used WMIC and vssadmin manually, and Scattered Spider (G1015) stopped the Volume Shadow Copy service.
- Hybrid and cloud ransomware: ATT&CK records Storm-0501 (G1053) deleting snapshots, restore points, storage accounts and backup services to prevent restoration. See the Cloud Ransomware Guide.
- Object storage: in the Codefinger S3 ransomware campaign, attackers with stolen AWS keys re-encrypted objects with customer-provided keys and used the S3 lifecycle API to schedule them for deletion within seven days, removing the window for recovery.
- Virtualisation: ESXi ransomware deletes snapshots before encrypting datastores. See the ESXi Ransomware Playbook.
- Destructive attacks: wipers use the same step to make sure systems cannot be rolled back. See the Wiper Malware Guide.
Impact
T1490 does not encrypt or destroy anything the business uses directly, but it decides how bad the next step is. With shadow copies and local backups gone, a single encrypted file server becomes a full restore from offline media; with cloud snapshots and versions gone, there may be nothing to restore at all. It also lengthens outages, increases pressure to pay, and destroys forensic artefacts held in snapshots. In cloud and hypervisor estates, one compromised administrator or backup credential can remove recovery points for thousands of workloads in minutes.
Detection
MITRE's v20 detection strategy for T1490 looks for process chains using native utilities with arguments that delete shadow copies, disable recovery or remove backup catalogues, plus platform equivalents. The CISA #StopRansomware Guide gives the same hunting list for Windows: anomalous use of bcdedit.exe, fsutil.exe (journal deletion), vssadmin.exe, wbadmin.exe and wmic.exe against shadow copies or shadow storage.
| Behaviour | Log source | Rule logic | Response |
|---|---|---|---|
| Shadow copy deletion or forced shrink | 4688 with command line or Sysmon Event ID 1; EDR | vssadmin, wmic, diskshadow or PowerShell WMI calls with delete or resize-shadowstorage arguments | Isolate host automatically; check for the same on other hosts in the last hour |
| Backup catalogue deletion | Process creation; Microsoft-Windows-Backup event log | wbadmin with delete arguments on a host that is not a backup server under change | Isolate; protect backup servers |
| Recovery disabled | Process creation; System log | bcdedit setting recovery off or ignoring boot failures; reagentc disabling WinRE | Isolate; record settings for restoration |
| VSS service stopped or disabled | System log service control events, Sysmon registry events | Volume Shadow Copy service set to disabled or stopped outside maintenance | Investigate the account responsible |
| Snapshot removal on ESXi | ESXi hostd logs forwarded to SIEM | Bulk snapshot removal or deletion of snapshot files from the shell | Disable shell, isolate management network. See the ESXi Logs Detection Guide |
| Cloud recovery tampering | Cloud audit logs (for example AWS CloudTrail) | Disabling backup plans, bucket versioning or snapshot schedules, reducing retention, new lifecycle expiry rules, then bulk snapshot, volume or image deletion | Revoke the session or key; check vault locks held |
| Network device recovery removal | Device syslog, AAA command accounting | Erase or format of flash followed by reload in one privileged session | Preserve configuration backups; isolate management access |
Tuning: legitimate uses exist (backup software managing its own snapshots, imaging and OS deployment tools), so allow by parent process, signer and host group rather than by command. Volume matters: the same command on ten hosts within minutes is ransomware deployment and should page someone at any hour.
Metrics
- Detection coverage: percentage of Windows hosts where T1490 command-line rules are active and tested with a safe simulation each quarter.
- Mean time from first T1490 alert to host isolation (target minutes, automated).
- Percentage of critical workloads whose backups sit in an immutable, locked vault or object-locked storage.
- Percentage of critical systems with a successful restore test in the last quarter.
- Number of identities able to delete or shorten retention on production backups (trend down).
Prevention and hardening
| Control | ATT&CK mitigation | Specific setting |
|---|---|---|
| Offline and immutable backups | M1053 Data Backup | Backups stored off system, encrypted and restore-tested, as the CISA guide recommends; cloud object lock or delete protection |
| Locked backup vaults (AWS) | M1053, M1018 | AWS Backup Vault Lock in compliance mode: after a grace time of at least 3 days the lock is immutable and AWS Backup denies deletion or lifecycle changes by any user, including root; set MinRetentionDays |
| Immutable vaults (Azure) | M1053, M1018 | Azure Backup immutable vault in "Enabled and locked" state (irreversible), which blocks deleting recovery points before expiry and reducing retention |
| Restrict delete APIs | M1018 User Account Management | Organisation-level policies (for example AWS Service Control Policies) that deny backup, snapshot and image deletion except to a break-glass role |
| Block unneeded utilities | M1038 Execution Prevention | Application control blocking diskshadow.exe and similar tools where not required |
| Keep recovery enabled | M1028 Operating System Configuration | Ensure WinRE is enabled and alert on configuration drift; protect services involved in recovery |
| Separate backup identity | M1018 | Backup consoles and vaults use a separate identity domain, MFA and no production domain admin rights. See the Backup Identity Isolation Guide |
See the Backup Security and Immutable Backup Guide, the AWS Backup Security Guide and the Azure Backup and Recovery Services Vault Security Guide for platform depth.
Rollout order
- Deploy detections first: they are cheap and high-value. Start in alert mode for one to two weeks to find legitimate backup and imaging tools, add allow conditions by signer and parent, then switch to automated isolation.
- Test safely: validate the rules with a controlled simulation on an isolated test host that lists or creates and deletes a test shadow copy, never on production data.
- Lock critical vaults: enable immutability in the reversible state first (governance mode in AWS, "Enabled" in Azure) to observe the effect on jobs and retention, then move to compliance mode or the locked state. Locking is a one-way door: review retention values before committing, because there is no rollback.
- Restrict deletion rights: apply organisation-level deny policies for backup deletion, with a documented break-glass role.
- Prove recovery: schedule restore tests for critical systems and record time to restore. See the Restore Testing and Recovery Assurance Guide.
Small versus large organisations: a small business gets most of the protection from an offline or immutable copy of critical data, a managed EDR with ransomware behaviour rules, and no shared admin accounts. Large estates need locked vaults per account and subscription, organisation-wide deny policies and SOC automation that isolates hosts in minutes, because deployment of T1490 across hundreds of hosts can happen faster than a human can respond.
Where identity fits
Every T1490 action runs under an identity with power over recovery: a domain or local administrator, a backup service account, a hypervisor root login or a cloud key with delete permissions. Codefinger used stolen long-term AWS access keys; Storm-0501 used privileged hybrid identities. Replace long-term keys with short-lived role credentials, give backup service accounts write but not delete rights where the platform allows, and alert when any identity that has never touched backup configuration starts changing it. See the Service Account Security Guide.
Lessons from incidents
- Lifecycle rules are a recovery attack: Codefinger did not need to delete anything immediately; scheduling deletion through the lifecycle API was enough to put a deadline on recovery. Alert on new or changed expiry rules on buckets holding business data, and keep versioning with protected backups so overwritten objects can be recovered.
- Cloud control planes need the same tripwire as endpoints: Storm-0501's deletion of snapshots, restore points and backup services happened through administrative access, not malware, so endpoint rules never fire. The equivalent detection is in cloud audit logs, and the equivalent prevention is a vault lock that the attacker's administrator identity cannot lift.
- The step is late but not last: in most ransomware cases T1490 comes minutes before encryption. Treat it as the final chance for automated containment and pre-authorise the SOC to isolate hosts and revoke sessions without waiting for approval.
Common failure patterns
- Treating snapshots as backups: relying on shadow copies or hypervisor snapshots that the attacker deletes in the same session.
- Backups in the production domain: backup servers joined to the same Active Directory, administered by the same domain admins the attacker already holds.
- Alert without action: a shadow copy deletion alert that waits in a queue overnight while encryption runs.
- Governance mode forever: leaving vault locks in a mode any sufficiently privileged identity can remove, which is the identity the attacker targets.
- Never restoring: immutable backups that have never been restored at scale, so the first real test is during the incident.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | PR.DS-11, DE.CM-09, RC.RP-03 | Backups created, protected, maintained and tested; runtime monitoring; integrity of backups verified before restoration |
| NIST SP 800-53 Rev. 5 | CP-9, CP-9(8), CP-6, CP-10, AC-6, SI-4 | System backup, cryptographic protection of backups, alternate storage site, recovery and reconstitution, least privilege, monitoring |
| CIS Controls v8.1 | Control 11, Safeguards 11.3 and 11.4 | Protect recovery data; maintain an isolated instance of recovery data |
| ISO/IEC 27001:2022 Annex A | A.8.13, A.5.30 | Information backup; ICT readiness for business continuity |
| MITRE D3FEND | D3-RF Restore File, D3-RDI Restore Disk Image | Restoration countermeasures that T1490 tries to deny |
Practitioner checklist
- Deploy command-line detections for vssadmin, wmic, wbadmin, bcdedit, reagentc and diskshadow recovery tampering, with automated isolation.
- Forward ESXi, cloud audit and network device logs and add the platform-specific T1490 rules.
- Keep at least one copy of critical data offline or in an immutable, locked vault.
- Deny backup, snapshot and image deletion by organisation policy except to a break-glass role.
- Separate backup administration identities from production administration.
- Replace long-term cloud access keys that can modify storage or backups.
- Restore-test critical systems at least quarterly and record time to restore.
- Rehearse the response to a multi-host shadow copy deletion alert.
Standards and references
- MITRE ATT&CK: Inhibit System Recovery (T1490)
- CISA: #StopRansomware Guide
- Microsoft Learn: vssadmin
- AWS: AWS Backup Vault Lock
- Microsoft Learn: Immutable vault for Azure Backup
- NIST: SP 800-53 Rev. 5
- NIST: Cybersecurity Framework 2.0
Related NHI Mgmt Group resources: Ransomware Prevention Guide · ATT&CK Impact Tactic Guide · Ransomware Response and Recovery Guide · Cyber Recovery Vault Guide