Tiering administrator accounts: the tier model

Why it is dangerous when a domain admin signs in to a workstation, and how the tier model separates the management of domain controllers and machines.

Topic
IT security
Author
ARLITECH expert team
Published

A domain admin reads email on their own laptop, then, from the same machine and with the same account, opens a Remote Desktop session to a domain controller. It is convenient, and on many networks nothing goes wrong for years. But if a malicious attachment runs on that laptop just once and the attacker gains local administrator rights, they can read from the machine's memory the credentials that can manage the entire domain.

The tier model gives a rule for this situation. It sorts systems into tiers by how much damage their takeover would cause, and it forbids an account from a higher tier to sign in to a lower-tier machine. Below we go through why a sign-in leaves traces, how the tiers are structured, which Windows settings enforce the rule, and in what order to introduce it.

Where an administrator sign-in leaves traces

In Windows, the logon type decides whether reusable credentials stay on the machine. With an interactive logon, meaning at the console or over Remote Desktop, the Local Security Authority subsystem process (LSASS) keeps the user's NTLM hash and Kerberos tickets in memory for the length of the session, so the user is not asked for the password again when accessing network resources. The same happens with a program started through RunAs. If a service or scheduled task runs under a domain account, its password is stored on the machine for good.

With a network logon, for example when someone opens a file share or a remote management console connects, reusable credentials usually do not reach the target machine. There are two exceptions: a connection authenticated with CredSSP, which forwards the password as well, and a target machine configured for unconstrained Kerberos delegation, which also receives the user's ticket-granting ticket (TGT). On top of that, the machine caches verifiers from earlier domain logons so it can let the user in without a domain controller. They cannot be used to authenticate directly, but they can be cracked offline.

Anyone with local administrator rights on a machine can read all of this. With an extracted hash (pass-the-hash) or ticket (pass-the-ticket), they can authenticate to other machines without knowing the password. That makes the domain controller reachable through any workstation a domain admin has recently signed in to.

The core idea of the model follows from this: a system is only as secure as the machine it is managed from and the account it is managed with. Microsoft's documentation calls this the clean source principle.

The three tiers: domain controllers, servers, workstations

The classic model separates three tiers. The number reflects the value of the access: taking over a Tier 0 system means taking over the whole environment.

TierWhat it containsTypical accounts
Tier 0Domain controllers, Active Directory itself, and everything that can control themMembers of Domain Admins, Enterprise Admins, Schema Admins, and the built-in operator groups
Tier 1Member servers, applications, databases, file serversServer and application administrators
Tier 2Workstations, laptops, user devicesHelp desk and workstation administrators

Two rules hold the tiers together. An account may control its own tier and the tiers below it, but it may not sign in to a lower-tier machine. Control cannot flow upward either: a workstation administrator cannot get rights on a server, and an agent running on a server cannot send commands to a domain controller.

Where Tier 0 ends

The hardest part of the design is deciding exactly which systems belong to Tier 0. Beyond the domain controllers, it includes every system that has indirect control over the domain: the certificate authority (AD CS), the federation server (AD FS), the server that synchronizes with Entra ID, the virtualization hosts running the domain controllers and their management interface, the system that backs up the domain controllers, and every management agent that can install software on a domain controller. Whoever administers the virtualization platform can also reach the domain controller's virtual disk, so in practice they hold Tier 0 rights.

The same goes for permissions. An account is Tier 0 even if it is not a member of Domain Admins, as long as it can reset the password of one of the members or has write access to a Group Policy object linked to the domain controllers. The article on Active Directory attack paths covers these hidden privilege chains in more detail.

Separate admin accounts and privileged access workstations

In practice, the rule means an administrator ends up with several accounts. The everyday account is for email, browsing, and editing documents. Alongside it, they get a separate admin account for each tier, but only for the tiers they actually manage. The name should show the tier, for example through a prefix, so that an account in the wrong place stands out right away in the logs and group memberships. Admin accounts have no mailbox and are not used for browsing the internet.

A separate account does little if it is used from the same laptop that runs email. If the laptop is infected, the attacker can also log keystrokes and capture the Tier 0 account's password as it is typed. Managing Tier 0 therefore calls for a privileged access workstation (PAW): a hardened machine that cannot be used for email or open browsing and runs only the management tools.

A PAW can be a separate physical machine, or it can be virtualized on the same hardware. In that case the more trusted system should be the host: the PAW runs on the hardware and everyday work runs in a virtual machine. In the reverse setup, an infected host can also see the management virtual machine running on it.

A jump host does not solve the problem on its own. If the administrator connects to it from an infected workstation, the credentials leak before they ever reach the jump host. A jump host helps only when the machine used to reach it is as clean as its tier requires.

A PAM system can tighten the handling of privileged accounts further, for example with time-limited privileges and session recording. The article on PAM covers this.

Logon restrictions with Group Policy

Leaving tier separation to discipline is not enough, because one rushed night of troubleshooting will undo it. Windows enforces it technically through User Rights Assignment. The following deny rights can be set in Group Policy:

  • Deny log on locally: the account cannot sign in at the machine's console.
  • Deny log on through Remote Desktop Services: the account cannot sign in over Remote Desktop either.
  • Deny access to this computer from the network: the account cannot reach the machine over the network either.
  • Deny log on as a batch job and Deny log on as a service: no scheduled task or service can run under the account.

The policy linked to workstations excludes the Tier 0 and Tier 1 admin groups, and the one linked to member servers excludes the Tier 0 groups. This needs an organizational unit structure in which machines and accounts sit in separate OUs per tier, so each policy can be linked unambiguously. A deny right always wins over an allow right, so not even membership in a local administrators group overrides it.

Denying network access goes beyond the sign-in rule, since apart from the exceptions mentioned earlier, a network logon leaves no reusable credentials on the target machine. It is still worth setting, because it closes those exceptions too, and it means workstations and member servers are managed with the admin accounts of their own tier.

Authentication policies and silos

Group Policy makes its decision on the target machine. Authentication policies and authentication policy silos, available from the Windows Server 2012 R2 domain functional level, restrict on the domain controller side which machines an account can get Kerberos tickets from. The restriction then takes effect when the ticket is issued, even if a machine's policy has not been refreshed for some reason. Device-based restrictions also require client-side support for Kerberos armoring (FAST).

The article on Windows hardening covers further hardening settings for workstations and servers.

The Protected Users group and credential protection

The Protected Users security group has existed since Windows Server 2012 R2. Windows applies stricter authentication rules to its members without any further configuration:

  • they cannot authenticate with NTLM, only with Kerberos;
  • DES and RC4 encryption cannot be used in Kerberos preauthentication;
  • the account cannot take part in any form of delegation;
  • the ticket-granting ticket (TGT) is valid for four hours by default and cannot be renewed, after which the user has to authenticate again;
  • no cached domain logon data stays on the client, and CredSSP and WDigest do not store plaintext passwords.

The client-side protections require Windows 8.1 or Windows Server 2012 R2 or later, and the domain controller side protections require the Windows Server 2012 R2 domain functional level.

Service accounts and computer accounts do not belong in the group. They often need NTLM or delegation and stop working once they are added. Add admin accounts gradually, because an old application that only supports NTLM will no longer let them in. Because cached logons are no longer available, keep one emergency account outside the group so that someone can still sign in when no domain controller is reachable. For admin accounts that cannot be members of the group for some reason, at least turn on the “Account is sensitive and cannot be delegated” flag.

Protecting memory

Credential Guard uses virtualization-based security to isolate NTLM hashes, Kerberos ticket-granting tickets (TGTs), and stored domain credentials, so even a program running with local administrator rights cannot read them directly. LSA protection runs the LSASS process as a protected process, which makes it harder for other processes to read its memory. For Remote Desktop connections, Remote Credential Guard and Restricted Admin mode keep the credentials from being passed to the target machine. In Restricted Admin mode, however, the session runs on the target machine under that machine's local identity, and Microsoft's documentation lists this as a pass-the-hash risk toward the target machine.

These techniques make reading credentials from memory harder. On a compromised machine an attacker can still capture a password as it is typed, so tier separation is needed alongside them.

A step-by-step rollout, from the tier model to the Enterprise Access Model

You cannot introduce the tier model in a working domain over a weekend. It touches every service account, scheduled task, and established workflow. Start with Tier 0, because its accounts are worth the most to an attacker and it has the fewest accounts.

  1. Tier 0 inventory. Which machines, accounts, groups, and services can control the domain, including nested group memberships and delegated rights.
  2. Separate accounts and a PAW for domain admins. Remove Tier 0 rights from the old all-purpose accounts. Only the dedicated accounts may stay in Tier 0 groups.
  3. Monitoring before you deny. Collecting 4624 security events shows where Tier 0 accounts sign in today. Logon type 2 indicates a local sign-in and type 10 a Remote Desktop sign-in. Anything that shows up here would cause an outage once the deny rules are in place, so sort it out first.
  4. Deny policies and Protected Users. Link the deny policies to workstations first, then to member servers. At the same time, add the Tier 0 admin accounts to the Protected Users group.
  5. Tier 1 the same way. Separate server admin accounts, deny rules on workstations, and local administrator passwords managed with LAPS, so that taking over one machine gives no access to the others.
Quick self-test: list the members of the Domain Admins group and check, for each one, which machine their daily work is tied to. If a member works with these rights from their own laptop, alongside their everyday account, that is where your tier model rollout begins.

The Enterprise Access Model

Microsoft's documentation now lists the tier model as a legacy approach, and its successor is the Enterprise Access Model. The tier principle survives in it, with a wider scope. Tier 0 corresponds to the control plane, which covers Entra ID and the cloud management interfaces as well as the on-premises domain. The management plane and the data and workload plane take the place of Tiers 1 and 2, and the model deals separately with the paths of user access and application access.

In a hybrid environment, this means a cloud Global Administrator is as valuable a target as a domain admin, and the same rules apply: a separate account, a protected device, a narrow role. The article on Microsoft 365 tenant security settings covers admin accounts on the cloud side. In on-premises Active Directory, the rules of the tier model still apply.

Related service: Active Directory review and hardening →

Contact

Ready to secure your critical systems?

Our experts will assess where your security stands and build a roadmap that fits your organization.

ARLITECH's office in Gödöllő

Visit us

You are welcome to visit our Gödöllő office in person.

Address
Centrum Office Building, 2100 Gödöllő, Dózsa György út 13, 3rd floor, office 309
Office hours
Mon–Fri, 9 a.m. to 5 p.m.
Get directions (opens in a new tab)