Deleting backups in a ransomware attack usually doesn’t require a complex exploit. The attacker becomes domain admin, signs in to the domain-joined backup server and deletes jobs and recovery points from the console. Five minutes’ work. This article explains ways to separate backup infrastructure from production identities: workgroup, separate management forest and local account models, console MFA, role separation, separating hypervisor and storage credentials, and management network design. The intended readers are system and security teams responsible for backup architecture and Active Directory.
Why backup infrastructure is targeted
The attacker wants to get paid; the organization being able to recover from backup ruins that plan. That’s why backup infrastructure is discovered and neutralized before the encryption stage. The typical moves are well known: disabling backup jobs, shortening the retention period to trigger cleanup of old recovery points, removing repositories from the console, changing the encryption password and stealing the credentials stored on the backup server. The backup server also holds credentials that reach hypervisors, storage and the inside of virtual machines. In other words, it’s not just a target but a springboard for lateral movement.
Identity design should answer two questions: How does someone who takes over production reach the backups? How does someone who takes over the backup infrastructure reach production? In the right design, both paths have to pass through separate identity and separate network boundaries.
The risk of domain membership
If the backup server is a member of the production domain, the following groups and mechanisms are effectively administrators on the server:
- The Domain Admins group is automatically added to the local Administrators group when the server joins the domain; Enterprise Admins also reach the same point with their forest-wide privileges.
- Anyone who can link a Group Policy to the OU containing the server can run code through a startup script or scheduled task.
- Service accounts of software distribution and endpoint management tools can run commands on the server.
- If console access has been granted to a domain account, every administrator who can reset that account’s password can get into the console too.
A domain-joined backup server is protected only as well as the domain’s weakest administrator account. If Active Directory has fallen, the backup server should be considered fallen too. A justification we often see is joining the backup server to the domain “to make management easy”; that ease applies to the attacker too. The administrative tiering approach in Active Directory is explained in the Active Directory tier model article; in this model, the backup server sits in the same, most critical tier as the domain controllers.
Identity models: workgroup, separate forest, local accounts
There are three common ways to separate backup infrastructure from production identity. Which fits depends on the size of the infrastructure and the team’s management capacity.
| Model | Suitable environment | Advantage | Limitation |
|---|---|---|---|
| Workgroup + local accounts | A single backup server, few components | Simple; not directly affected even if the domain is compromised | No Group Policy; updates and account management by hand |
| A separate management forest (no trust) | Many proxies, repositories and administrators | Central accounts, policy and auditing; MFA integration becomes easier | Requires additional DCs and separate patching and monitoring processes |
| Production domain + separate OU and accounts | Transition period | Easy to set up | Provides no protection against a domain admin |
If you set up a separate management forest, don’t create a trust with the production forest. If a trust seems necessary, its direction should be production trusting the management forest; the reverse means identities in production can reach the backup infrastructure. Backup usually requires no trust at all, because the backup software connects to protected systems with its own stored credentials.
The workgroup model requires discipline. Create a separate local account for each administrator, don’t use a shared “admin” account, and put the built-in Administrator account’s password in the vault and take it out of everyday use. Apply password, account lockout and audit policies with a documented baseline configuration and record changes. Without Group Policy, you are the record.
MFA and role separation in the backup console
Separating the server from the domain is the first step; the console itself also needs protecting. Taking Veeam Backup & Replication as an example, version 12 brought MFA with time-based one-time passwords (TOTP) for console users, and version 12.1 brought four-eyes authorization for critical operations. Verify the scope and limitations of these features in your version (for example, MFA support for group accounts) in the vendor’s documentation; look for equivalent controls in other backup products.
| Check | Recommended setting | Risk it closes |
|---|---|---|
| Console MFA | Mandatory for all users | Console sign-in with a stolen password |
| Role separation | Administrator, operator and restore operator separated | Unnecessary use of full rights in daily work |
| Four-eyes approval | On for deleting backups, repositories and users | Mass deletion with a single account |
| Session timeout | Short period, automatic lock | A console session left open |
| Number of local administrators | As few as possible; named accounts | Untraceable actions with a shared account |
The rule for role separation is simple: the person who monitors and restarts daily jobs isn’t given the full administrator role. Most restore requests are handled with the restore operator role. The number of accounts in the full administrator role should be limited to two or three, and these accounts shouldn’t be used for everyday sessions. When you turn on four-eyes approval, make sure the second administrator who’ll give approval is always reachable. Otherwise urgent operations get stuck.
Separating hypervisor and storage credentials
The backup server stores credentials to connect to protected systems. The scope of these credentials determines how far an attacker gets when the server falls.
- vCenter: Instead of a domain account, create a separate account in vCenter’s local SSO domain (for example vsphere.local) and assign a custom role containing only the privileges documented by the backup product. Don’t give it the Administrator role; that’s the excess privilege we often find.
- Hyper-V: If the hosts are members of the production domain, the backup account is usually a domain account too; make sure this account is a local administrator only on the Hyper-V hosts and that its right to sign in interactively is restricted. The hosts themselves should be managed at the most critical tier.
- Guest OS processing: Application-aware backup requires administrator rights inside virtual machines. Use separate accounts for server groups; since the account used for domain controllers will have domain-level privileges, make sure it’s defined only for DCs.
- Storage arrays and object storage: The delete rights of accounts used for snapshot integration and object storage should be minimized; administrator accounts that can change the immutability setting should never be stored on the backup server.
- Hardened repository: No permanent credentials are kept on the backup server; installation details are in the Veeam hardened repository setup article.
For each credential, answer this question in writing: If this account is compromised, what can be done on which systems? Accounts whose answer is “the whole domain” or “all storage” go onto the list of most critical assets and are subject to rotation, monitoring and vault storage rules. A privileged access management (PAM) approach can be used to store privileged accounts in a vault and record sessions.
The management network and access path
On the enterprise cybersecurity side, identity separation is incomplete without network separation. A backup server separated from the domain but reached over RDP from the user network is still compromised with a stolen local password. The structure we recommend:
- The backup server, proxies and repositories in a separate management VLAN or segment.
- Management access to this segment only from privileged access workstations or a jump server reserved for this purpose; access to RDP, SMB and console ports from the user network is blocked.
- Ports needed for backup traffic are open per source and destination; there’s no “allow everywhere from the management network” rule.
- The backup server’s direct internet egress is off; the addresses needed for updates and license validation are defined.
- Remote management cards are on a separate out-of-band management network.
If the jump server is also a member of the production domain, the same risk comes back through the back door. The jump point should be a member of the management forest or in a workgroup and protected with MFA. The primary source for the privileged access workstation concept is Microsoft’s privileged access strategy documentation.
Implementation checklist
The list was prepared to proceed in order when moving an existing environment to an isolated structure. Run a test job after each step so backup jobs aren’t interrupted during the transition. A backup night skipped during the transition comes back to haunt you on the worst night.
- An inventory has been made of which domain accounts and groups the backup server, proxies and repositories depend on.
- The backup configuration has been backed up; the encryption password is outside the domain, in the vault.
- A new identity model has been chosen (workgroup or a separate management forest); there’s no trust with production.
- Console users have been moved to local or management forest accounts; domain accounts’ console access has been removed.
- Console MFA is on for all users; the number of full administrators is at a minimum.
- Four-eyes approval is enabled for delete and repository change operations.
- vCenter, Hyper-V, guest OS and storage accounts have been reviewed; each has been limited to least privilege.
- Backup components are in the management segment; access from the user network is blocked.
- Events on the backup server (sign-ins, job deletions, setting changes) flow into the central log system.
- At least one immutable or offline copy is protected with identities independent of the backup console.
Isolation makes an attack harder; it doesn’t replace detection. How to monitor suspicious operations in the console is covered in the signs of attacks on backup infrastructure article. Adapting identity separation to your existing architecture is one of the main topics of our backup security and ransomware readiness work.
Frequently asked questions
If the backup server isn’t a domain member, won’t backups work?
They work. The backup software connects to the hypervisor, storage and inside virtual machines with its own stored credentials; the server doesn’t need to be a domain member. What changes is the management method: updates, local accounts and security settings are managed through a separate process instead of Group Policy.
How should I choose between a separate management forest and a workgroup?
If the backup infrastructure is just one or two servers, a workgroup and local accounts are simple and manageable. In environments with many proxies, repositories and administrators, a separate management forest is more sustainable for central management, MFA and audit logging. In both cases, don’t create a trust relationship with the production domain.
Is adding MFA to the backup console enough protection?
No. MFA protects console sign-in; but someone who is an administrator on the backup server’s operating system can directly interfere with the database, the configuration backup or the services. MFA becomes meaningful together with positioning outside the domain, role separation, four-eyes approval and an immutable repository.
Why are guest OS processing credentials risky?
Application-aware backup requires administrator rights inside virtual machines, and these credentials are stored on the backup server. If domain controllers are also in scope, the stored account has domain-level privileges. Someone who takes over the backup server reaches these credentials; that’s why the backup server should be protected at the most critical tier.