Defender for Identity: Deployment and Alert Assessment

Before installing the Defender for Identity sensor, check the server role, the supported sensor version and event logging. The guide separates the v2.x and v3.x requirements.
Technical summary
Technology
Microsoft Defender for Identity, Active Directory, Advanced Audit Policy, gMSA, Defender XDR, PowerShell
Topic
AD threat detection sensor deployment and alert triage
Level
Intermediate–advanced
Prerequisites
  • A license that includes Defender for Identity
  • Local administrator and GPO management rights on DCs
  • 443/TCP access from DCs to the cloud service
In this article8 sections

Defender for Identity is deployed, the sensors are green, the portal is quiet. This silence usually indicates not a clean environment but missing audit policies, a service account with wrong permissions or a DC without a sensor. Microsoft Defender for Identity is a cloud-based detection service that monitors attack behavior aimed at on-premises Active Directory with sensors installed on DCs and identity infrastructure servers. In this guide we explain the sensor requirements, audit configuration, the meaning of alert types, the triage order and security posture assessments, along with the points where deployments commonly get stuck.

What Defender for Identity sees and doesn’t see

The sensor collects data from three sources on the server it’s installed on: network traffic to and from the DC (Kerberos, NTLM, LDAP, DNS, SMB, RPC), Windows event logs, and object information obtained through directory queries. The data is processed in the cloud with behavioral models and appears in the Microsoft Defender XDR portal as alerts and incidents.

Knowing its limits is also part of the deployment. Defender for Identity doesn’t see process behavior on endpoints (that’s Defender for Endpoint’s area), doesn’t analyze Entra ID cloud sign-ins on its own, and never sees an attack going through a DC without a sensor. A DC it doesn’t see is a dark room. It partially reports static configuration weaknesses through security posture assessments; deep ACL and privilege path analysis remains a separate exercise, AD attack path analysis.

Sensor requirements and installation

Servers to install the sensor on:

  • All writable DCs and RODCs. Leaving a single DC out means the activity going through that DC isn’t visible; the attacker can choose which DC to talk to.
  • AD FS, AD CS (certificate authority) and Entra Connect servers. These are Tier 0 identity infrastructure and the targets of attacks specific to them.

Pre-installation checklist:

  • Version and resources: For supported Windows Server versions and the sensor’s hardware requirements, use Microsoft’s current requirements page and capacity planning tool; extra CPU and memory may be needed depending on the DC’s current load.
  • Connectivity: The sensor must reach the Defender for Identity cloud service over 443/TCP. If there is a proxy, it’s defined with installation parameters and excluded from SSL inspection.
  • Network name resolution (NNR): To map IP addresses to device names, the sensor uses NTLM over RPC (135/TCP), NetBIOS (137/UDP), RDP (3389/TCP, first packet only) and reverse DNS queries to clients. Internal firewalls must allow these connections from DCs to client networks; otherwise alerts show only the IP instead of the device name, and every triage starts with a DHCP log lookup.
  • Virtual DCs: For DCs on VMware, Microsoft has documented a compatibility note about the virtual NIC’s Large Send Offload setting; check it before installation.

Sensor v2.x (the classic sensor) is installed with the package downloaded from the portal and the workspace-specific access key. A silent installation example:

"Azure ATP Sensor Setup.exe" /quiet NetFrameworkCommandLineArguments="/q" AccessKey="<access-key-from-portal>"

Sensor v3.x is activated from the portal without installing a separate package. According to Microsoft’s deployment documentation, the server must run Windows Server 2019 or later, must have received the cumulative update specified in the documentation, and must not have a v2.x sensor on it; activation can be done on servers onboarded to Defender for Endpoint, or on eligible servers without Defender for Endpoint. On older operating systems, v2.x is used. Before deployment, check the current status of v3.x support on AD FS, AD CS and Entra Connect servers that aren’t DCs (documented as preview when this article was updated) in the Defender for Identity documentation.

Audit policies and the directory service account

A significant share of detections relies on Windows events. If the required Advanced Audit Policy settings aren’t applied to the DCs, the sensor looks healthy but some alerts are never generated. A healthy sensor doesn’t mean working detection. The main settings:

Sensor v3.x has an option to configure audit settings automatically. With sensor v2.x, or when this option is off, the settings are made manually or with the PowerShell module below.

Audit subcategory Value Example events
Credential Validation Success, Failure 4776
Computer Account Management Success, Failure 4741, 4743
Security Group Management Success, Failure 4728, 4729, 4732, 4733, 4756, 4757
User Account Management Success, Failure 4726
Directory Service Access Success, Failure 4662
Directory Service Changes Success, Failure 5136
Security System Extension Success, Failure 7045

In addition, object auditing (SACL) must be defined on the domain root for 4662 events to be generated, and NTLM auditing must be turned on on DCs for NTLM detections (8004 events; details in the NTLM auditing and restriction article). AD FS, AD CS and Entra Connect servers need additional audit settings.

Microsoft’s DefenderForIdentity PowerShell module reports the current configuration and can apply the required settings as a GPO. It’s the first tool to run before fiddling with GPOs by hand:

Install-Module DefenderForIdentity -Scope CurrentUser
# Status of the required audit configurations across the domain
New-MDIConfigurationReport -Path C:\mdi -Mode Domain
# Apply missing settings with a new GPO (review the report first)
Set-MDIConfiguration -Mode Domain -Configuration All

Directory service account (DSA): Sensor v2.x uses a directory service account for directory queries. The recommended type is a gMSA; its password is rotated automatically by AD. With sensor v3.x, no DSA or gMSA is configured; the sensor runs as LocalSystem. The DSA only reads; the action account that carries out response actions such as disabling an account or resetting a password is a separate account. The steps in this paragraph are for v2.x. The account must have read permission on all objects and on the Deleted Objects container; the computer accounts of the servers where the sensor is installed (in practice, a group containing them) must be added to the gMSA’s PrincipalsAllowedToRetrieveManagedPassword list. If a GPO restricts the “Log on as a service” right on sensor servers, the gMSA is added to that right too. A common mistake here: a DC whose computer account was newly added to the group can’t retrieve the gMSA password until it renews its Kerberos ticket; a restart or ticket purge is needed. For lateral movement path calculation, the account needs to be able to make SAM-R queries on endpoints; this is done by adding the DSA to the “Network access: Restrict clients allowed to make remote calls to SAM” policy, and it should be applied only to the necessary scope of computers.

Alert types and what they mean

Alerts are grouped by the stages of the attack chain. Common types and their first interpretation on the defensive side:

Alert Stage First interpretation
Security principal reconnaissance (LDAP) Discovery Bulk LDAP query; authorized analysis tools trigger it too
User and IP address reconnaissance (SMB) Discovery An attempt to enumerate sessions on a DC
Suspected Kerberos SPN exposure Credential access Possible Kerberoasting; review the targeted service accounts
Suspected AS-REP Roasting attack Credential access Requests to accounts with pre-authentication turned off
Suspected brute-force attack (Kerberos, NTLM, LDAP) Credential access Password guessing or password spray
Suspected identity theft (pass-the-ticket) / overpass-the-hash Lateral movement Use of a stolen ticket or hash
Suspected DCSync attack Credential access A replication request from a non-DC source; high priority
Suspected Golden Ticket usage Persistence A forged TGT; suggests krbtgt has been compromised
Suspicious additions to sensitive groups Privilege escalation An unexpected member in a privileged group

For background on Kerberos-based alerts, see the Kerberoasting and AS-REP roasting article.

Alert triage and investigation steps

Following the same order for every alert improves both speed and consistency of decisions. Even if the analyst changes, the outcome shouldn’t:

  1. Source: Which device and which account did the activity come from? Is the source device a management server, a scanner or an authorized analysis tool?
  2. Target: Which DC, which accounts or groups were affected? Is the target Tier 0?
  3. Timeline: Earlier and later activity from the same source; the link with endpoint and email alerts in the Defender XDR incident view.
  4. Verification: Find the related process on the source device with the Defender for Endpoint timeline; does it match a change record or scheduled job?
  5. Classification and action: If it’s a true positive, disable the account, reset the password, isolate the device; if it’s benign but expected activity (benign true positive), close it with the reasoning and define an exclusion if needed.

Keep exclusions narrow. Excluding a specific scanner server from a specific detection is acceptable; removing an entire subnet from all reconnaissance detections means the attacker stays invisible from that subnet. A broad exclusion is a blind spot.

Security posture assessments

Defender for Identity also derives configuration weaknesses from the data the sensors collect and presents them as recommendations in Microsoft Secure Score. Notable assessments:

  • Insecure Kerberos delegation (accounts and computers with unconstrained delegation).
  • Services sending clear-text credentials over the network (e.g. simple LDAP bind).
  • Long-unused accounts in sensitive groups.
  • Use of legacy protocols and weak encryption (NTLMv1, RC4).
  • The Print Spooler service running on DCs.
  • Insecure account attributes (password never expires, pre-authentication turned off, password stored with reversible encryption).
  • Unmonitored DCs (no sensor installed).

These recommendations aren’t a fixed checklist; they’re a list of the real objects in your environment. Assigning an owner to each and tracking the remediation joins the detection layer with the prevention layer in enterprise cybersecurity. A recommendation without an owner stays open forever.

Common deployment mistakes

  • Incomplete sensor coverage: Installing on only a few DCs, forgetting AD CS and Entra Connect servers. Forgetting to install the sensor on a DC added later has the same result; write sensor installation into the DC promotion procedure.
  • Missing audit policies: The Default Domain Controllers Policy doesn’t configure all the advanced audit subcategories Defender for Identity requires; check sensor health alerts regularly.
  • NNR blocked on the firewall: Device matching fails in alerts.
  • SSL inspection: The sensor’s cloud connection is broken by an intercepting proxy.
  • Nobody watching health alerts: Nobody notices when a sensor stops or the event flow is cut. Route health notifications to the SOC queue.

Monitoring sensor alerts 24/7 and correlating them with other sources is the subject of the SOC and managed security model; pre-deployment audit configuration and remediation of posture findings can be planned together with an Active Directory security audit.

Frequently asked questions

Does the Defender for Identity sensor affect DC performance?

Because the sensor processes traffic and events, it uses extra CPU and memory on the DC. To keep the impact acceptable, measure the current load with Microsoft’s capacity planning tool before deployment and increase DC resources if needed.

Is it enough to install the sensor on only some DCs?

No. The attacker can choose which DC to talk to; activity going through a DC without a sensor is invisible. Installing on all DCs and on AD FS, AD CS and Entra Connect servers is recommended.

This question applies to sensor v2.x; v3.x doesn’t use a directory service account. A gMSA’s password is rotated automatically and with complexity by AD, can’t be used for interactive sign-in, and can be retrieved only by permitted servers. It carries lower risk than a service account with a fixed password.

What should you do if an authorized vulnerability scan generates alerts?

Classify the alert as benign but expected activity and record the reasoning. If the scan is regular, define a narrow exclusion only for the scanner device and the relevant detection type; avoid broad subnet exclusions.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Cybersecurity

Firewall, VPN, endpoint security and threat prevention. We reduce the attack surface and measure and report the risks.

Knowledge Center

Other articles