A Domain Admin password is rarely stolen from a domain controller. It usually comes out of an ordinary workstation where someone signed in with that account, or out of an application server’s memory. The Active Directory tier model is an access architecture built so that administrator credentials never land on these lower-security systems in the first place. Below we explain how the Tier 0/1/2 boundary is drawn, the account structure, logon restrictions with GPO, PAWs and an implementation order that doesn’t disrupt current operations. Once the security changes are in place, handle directory health and periodic maintenance within Active Directory and server management.
Credential theft and the logic of the tier model
In Windows, operations such as an interactive session, an RDP connection, starting a process with “Run as” and storing a password in a scheduled task leave derived forms of the credential (NT hash, Kerberos tickets) in the target system’s memory. An attacker who gains local administrator rights on that system reads them and moves on to other systems with pass-the-hash or pass-the-ticket. The help desk RDPing to a user’s computer with a Domain Admin account can carry an ordinary piece of malware on that machine to the entire domain. One RDP is enough.
The tier model rests on two rules:
- The control rule: Everything that controls an asset is in the same tier as that asset. The server that backs up a domain controller is Tier 0 too.
- The logon rule: A higher-tier account doesn’t sign in on a lower-tier system. A Tier 0 account leaves no credentials on a workstation; a lower-tier account can’t manage a higher-tier system either.
The model isn’t a product; it’s a working discipline at the foundation of enterprise cybersecurity. Technical controls such as GPOs, authentication policies and separate accounts exist to make stepping outside this discipline harder.
Identifying Tier 0, 1 and 2 assets
A common mistake is counting Tier 0 as just the DCs. When you apply the control rule honestly, the list gets longer:
| Tier | Example assets | Why this tier |
|---|---|---|
| Tier 0 | DCs, AD CS, AD FS, the Entra Connect server | Produces or synchronizes identity itself |
| Tier 0 | The hypervisor hosting the DCs, the backup system that backs up the DCs | Can access the DC’s disk and memory |
| Tier 0 | Management tools that deploy agents to DCs, the PAM vault | Can run code on the DC |
| Tier 1 | Application, database, file and email servers | Hosts business data |
| Tier 2 | Workstations, laptops, printers, user support tools | Open to user use, exposed to the internet |
The practical test is this question: “Can this system’s administrator run code on a DC or read the DC’s disk?” If the answer is yes, the system is either Tier 0 or its relationship with the DC must be cut. For example, if a single hypervisor cluster carrying all servers also hosts the DCs, there are two ways: the cluster is counted as Tier 0 and its management is narrowed, or the DCs are moved to a separate management plane. There’s no middle ground. Similar decisions for backup systems are covered in the identity isolation in the backup infrastructure article.
Administrator account structure and naming
Each administrator uses a separate account for every tier they need, and does email and internet work with their unprivileged standard account. Naming should show the account’s tier at first glance to both people and scripts.
| Calculation | Example name | Where it’s used |
|---|---|---|
| Standard user | ayilmaz | Email, internet, business applications |
| Tier 2 administrator | t2-ayilmaz | Workstation support |
| Tier 1 administrator | t1-ayilmaz | Server administration |
| Tier 0 administrator | t0-ayilmaz | AD and Tier 0 systems, only from a PAW |
Accounts are kept in separate OUs by tier, and each tier has its own administrator group. Tier 0 accounts have no mailbox, aren’t synchronized to Entra ID and are added to the Protected Users group. For Protected Users members, NTLM, DES/RC4 in Kerberos pre-authentication and delegation can’t be used; the TGT lifetime is limited to four hours and credentials aren’t cached. These restrictions can break old applications; try them with a few accounts first, then roll out.
The first step is seeing the current state: list the privileged groups, resolving nested memberships. The length of this list is the first surprise in most environments.
Import-Module ActiveDirectory
$groups = 'Domain Admins','Enterprise Admins','Schema Admins',
'Administrators','Account Operators','Backup Operators',
'Server Operators','Print Operators'
$result = foreach ($g in $groups) {
Get-ADGroupMember -Identity $g -Recursive |
Where-Object objectClass -eq 'user' |
Get-ADUser -Properties LastLogonDate, PasswordLastSet, Enabled |
Select-Object @{n='Group';e={$g}}, SamAccountName, Enabled,
LastLogonDate, PasswordLastSet
}
$result | Sort-Object Group, SamAccountName | Format-Table -AutoSize
Built-in groups such as Backup Operators, Server Operators and Account Operators grant indirect Tier 0 rights on the DC. The goal is for these groups to be empty.
Enforcing logon restrictions with GPO
The logon rule is technically enforced with User Rights Assignment settings. The settings are under Computer Configuration > Policies > Windows Settings > Security Settings > Local Policies > User Rights Assignment and are applied with separate GPOs linked to the tier OUs.
| Target systems | Groups to deny | Settings |
|---|---|---|
| Tier 2 workstations | Tier 0 and Tier 1 administrators | All deny rights |
| Tier 1 servers | Tier 0 administrators | All deny rights |
| Tier 1 servers | Tier 2 administrators | No administrative rights granted; deny optional |
| Tier 0 systems | Tier 1 and Tier 2 administrators | Interactive and RDP deny |
“All deny rights” means these five settings: Deny log on locally, Deny log on through Remote Desktop Services, Deny access to this computer from the network, Deny log on as a batch job and Deny log on as a service. Be careful when applying network access denial to DCs: users need network access to the DC for Group Policy and authentication; in the DC GPO, list only the administrator groups that really need to stay out.
For tighter control, there are Authentication Policies and Authentication Policy Silos, available from the Windows Server 2012 R2 domain functional level: Tier 0 accounts’ Kerberos TGTs can be obtained only from specific Tier 0 computers. Using the credential on another machine is blocked on the DC side, and bypassing this is harder than bypassing a GPO; in return, it requires Kerberos armoring (FAST) support and careful testing.
When connecting to Tier 1 servers with RDP, consider Remote Credential Guard or Restricted Admin mode so the credential doesn’t stay on the target. Local administrator accounts having a different password on each machine is also a complementary control that cuts lateral movement; the steps are in the migrating to Windows LAPS article.
Privileged access workstation (PAW)
The device where a Tier 0 account’s password is entered is Tier 0 too. An administrator who RDPs with a Tier 0 account from their everyday laptop undoes all the GPOs you’ve set up in one move. A PAW is a hardened device used only for administrative work:
- There’s no email client, general web browser or office applications; internet access is limited to admin portals.
- Only Tier 0 administrators can sign in; standard user accounts are denied.
- Credential Guard, Secure Boot, BitLocker and application control (WDAC/AppLocker) are enabled.
- The device is managed in the Tier 0 OU; Tier 2 management tools can’t install agents on it.
If a separate physical device can’t be given to each administrator, central, hardened jump servers can be used for Tier 0 access. But the device that connects to the jump server is still a question. The jump server is just an additional hop; it doesn’t remove the risk of the source device.
The relationship with the enterprise access model
Microsoft updated the classic tier model and the ESAE (“red forest”) recommendation with a broader framework, the enterprise access model. This model also covers hybrid and cloud environments; it uses the concepts of control plane, management plane and data/workload plane. For on-premises AD, Tier 0 roughly corresponds to the control plane; roles in Entra ID such as Global Administrator and Privileged Role Administrator are on the same plane.
In practice this means: on-premises Tier 0 and cloud identity administration can’t be thought of separately. An attacker who compromises the Entra Connect server can affect identity both in the cloud and on premises. This server is Tier 0. Just-in-time authorization of privileged roles on the cloud side is described in the Entra ID PIM guide. If you’re looking for a general starting point, the Active Directory security controls article lists the basic controls to apply alongside the tier model.
Implementation order and common mistakes in the field
Order matters. Start with the step with the lowest risk of breakage:
- Inventory: Privileged group members, service accounts and Tier 0 candidate systems are listed; who manages each system is written down.
- Drawing the Tier 0 boundary: All systems that control the DCs are identified; if needed, the relationship of backup, hypervisor and management tools with the DCs is cut.
- Account separation: Tier 0 accounts are created, Domain Admins membership is moved to these accounts, and privileges are removed from personal accounts.
- PAW or jump server: Tier 0 access is reduced to a single controlled path.
- Deny GPOs: First Tier 0 accounts are denied on Tier 2 (the lowest risk of breakage), then on Tier 1 servers.
- Monitoring: 4624 logon events for Tier 0 accounts on non-Tier 0 systems raise alarms in the SIEM.
Common mistakes:
- Leaving service accounts in the Domain Admins group; because these accounts run on Tier 1 servers, they punch straight through the model.
- Counting backup and monitoring systems as Tier 1 and leaving them with full access to the DCs.
- Adding the Tier 1 administrator to the local Administrators group on all servers; separation by workload is needed within Tier 1 too.
- Opening new privileged groups for exceptions after the deny GPOs are done and not monitoring them. Over time, the exception becomes the rule itself.
To see where the model is actually breached, you need to look at the privilege paths formed by configurations combining; the method is in the AD attack path analysis article. For privileged access design, see the privileged access management page, and for assessing the current structure, the Active Directory security audit page.
Frequently asked questions
Is the tier model necessary in a small environment too?
Credential theft doesn’t care about the size of the environment. Even if a full PAW infrastructure can’t be set up, separate Tier 0 accounts, a GPO preventing Domain Admin accounts from signing in on workstations and cleaning up privileged groups are applied at low cost and close most of the risk.
Why shouldn’t Tier 0 accounts be synchronized to Entra ID?
A synchronized Tier 0 account becomes an attack surface on the cloud side too; an identity compromised in the cloud or a password reset flow can open a path to on-premises Tier 0. Keeping on-premises and cloud administrator accounts separate cuts the privilege crossover between the two environments.
Do deny logon GPOs break applications?
Not when applied to the right groups. However, if there are service accounts that are members of privileged groups, the services these accounts run may not start. Clean up the service accounts’ group memberships first, test the GPO in a pilot OU, then roll it out in stages.
Has the enterprise access model replaced the tier model?
The enterprise access model extends the tier model to cover hybrid and cloud environments. For on-premises Active Directory, tier separation and logon restrictions remain valid; the new model carries this separation to cloud identities and application access with the concepts of control plane, management plane and data/workload plane.