Before starting encryption, a ransomware operator looks for the backups. They log into the backup server’s console, delete the job history, clear the snapshots on the storage, then start encrypting. That’s why the sentence “we have backups” says nothing on its own. The question to ask is: once the attacker is a domain administrator, can the backups still be restored?
The five topics below describe the preparation needed to answer “yes” to that question. The order is deliberate: first the backup surviving, then the identity being separated, and the exercise last.
1. Put backups where the attacker can’t reach them
- Immutable copy: Storage or an object lock that can’t be deleted or changed for a defined period.
- Offline / air-gapped copy: Tape or media disconnected from the network.
- Different location: A copy that won’t be affected by a building- or region-level event.
The widely used 3-2-1 rule (3 copies, 2 different media, 1 off-site) is extended to 3-2-1-1-0 in the ransomware context: one copy immutable or offline, zero errors in verification.
In practice, immutability is achieved in a few ways: Object Lock on S3-compatible object storage (in compliance mode, the retention period can’t be shortened even by an administrator), a hardened Linux repository supported by the backup software, or the locked snapshot feature of a storage array. Whichever you choose, the retention period should cover the time an attacker can remain undetected in the environment. A copy whose lock lasts a few days doesn’t protect against an attacker who got in two weeks ago.
Common mistakes
- The backup target NAS being an SMB share mapped as a drive letter on the backup server. Encryption reaches this share too.
- Root access via SSH or a web interface left open on a repository marked “immutable”. If the lock doesn’t protect against the operating system administrator, it isn’t a lock.
- Counting hypervisor snapshots as backups. A snapshot is deleted with one click from the same management console.
- The cloud copy sitting in an account managed with the same administrator account as production.
2. Separate the identity of the backup infrastructure
If the backup server and storage are managed with the production domain’s administrator accounts, a compromised administrator account has also compromised the backups. Separate accounts, MFA and, if possible, management separate from the domain should be preferred for the backup infrastructure. Only the backup server should be able to access backup storage.
Concretely, this means: the backup server isn’t a member of the production domain, or it sits in a separate management forest. Console logins are done with local accounts or accounts in a separate directory, with MFA. Service accounts used for hypervisor and storage management are defined with least privilege; entering an account in the “Domain Admins” group as the backup job’s credentials is handing the key to the attacker.
The network side is part of this separation too. The backup network should be in a separate VLAN, and firewall rules should open only the ports needed for backup traffic. There should be no access to management interfaces from the user VLAN. A flat network makes the attacker’s job easier.
3. Test restores regularly
A successful backup job doesn’t mean a successful restore. For critical systems, a restore test should be done at least every three months, with the times measured and recorded. Test results make it possible to set realistic RTO (recovery time) and RPO (acceptable data loss) targets.
What makes the test meaningful is what you measure. Saying “the VM booted” isn’t enough. Did the application services come up, is the database consistent, can users sign in? These should be checked. Most backup software can automatically boot the restored machine in an isolated network and verify it with a script; use this feature and report the results.
A common problem is that the bandwidth calculation was never done. Downloading a ten-terabyte file server from a cloud copy can take days, depending on your line speed. If that time doesn’t match your RTO target, a local immutable copy is a must. An RTO given without doing the math is a guess.
4. Decide the recovery order in advance
| Priority | System | Why |
|---|---|---|
| 1 | Identity infrastructure (AD / DNS) | Most other systems depend on it; you must be sure it’s clean |
| 2 | Network and security components | Isolation and monitoring of the clean environment |
| 3 | Business-critical applications | ERP, production planning, finance |
| 4 | File and collaboration systems | Users getting back to work |
To determine a clean restore point, backups need to be scanned for malware and the start time of the attack needs to be estimated.
The topic that wastes the most time in the sequence is Active Directory. If the attacker established persistence in the domain (new administrator accounts, a malicious GPO, a KRBTGT key compromised for a golden ticket), a domain controller restored from yesterday’s backup brings back the same back door. That’s why AD is restored in an isolated network, privileged groups and GPOs are checked, and only then is it connected to production. Having read and rehearsed Microsoft’s forest recovery guidance in advance saves hours here.
Talk through dependencies with application owners as well. The ERP server depends on SQL, SQL on AD for authentication, and the license service perhaps on a separate server. If this chain hasn’t been put on paper, it gets solved by trial and error on the day of the incident.
5. Write the plan and exercise it
The recovery plan should include who makes decisions, which systems come back in which order, communication steps and external support contact details. A tabletop exercise is the fastest way to see whether the plan will work in a real incident.
The plan itself can be affected by the attack. A Word document on the file server can’t be read when the file server is encrypted. There should be a printed copy of the plan and a copy stored somewhere independent of the company infrastructure. The same applies to passwords: the emergency account credentials for the backup console and hypervisor shouldn’t be locked inside the password vault that gets encrypted.
In the exercise, these questions should get clear answers:
- Who decides to shut down systems or disconnect them from the network, and who is reached at midnight?
- If email and Teams aren’t working, which channel does the team use to communicate?
- At what stage do cyber insurance, legal and the incident response firm get involved?
- Which system comes back from which copy, and in how many hours?
Short list for first response
- Isolate affected systems from the network; consider preserving evidence before shutting them down.
- Verify that the backup infrastructure is reachable and intact, and restrict access to backup systems.
- Reset passwords and sessions of privileged accounts in a controlled way.
- Activate the incident response plan; assess legal notification obligations with your legal team.
The most common mistake in the first hour is shutting down and restarting all servers in a panic. Traces in memory are lost, and some encryption processes leave half-done files unusable. Isolation comes before shutting down. Shutting a switch port or disconnecting a virtual machine’s network adapter is enough and safer in most cases.
For a security assessment of your backup infrastructure, see our Backup Security and Ransomware Readiness page, and for the response plan, our Cyber Incident Response Readiness page; for backup setup and operation, see the Eretek System Support service.