Detect DCSync and ntds.dit Access

Goal

Detect both forms of bulk Active Directory credential extraction — DCSync (IDL_DRSGetNCChanges via the DRSUAPI RPC interface) and offline ntds.dit extraction (file-level access, VSS shadow-copy abuse, ntdsutil invocations) — with high enough fidelity to identify the source principal, the affected accounts, and the post-extraction window before forged tickets (Golden Ticket or Silver Ticket) can be used.

Assumptions

you collect Windows Security event logs from all Domain Controllers into a SIEM with ≥ 90 days retention
you have EDR coverage on every Domain Controller (Sysmon, Microsoft Defender for Identity, or equivalent)
you maintain an authoritative list of principals with DS-Replication-Get-Changes-All and related replication rights (audited monthly via BloodHound)
you can hot-disable a compromised account and initiate the twice-in-sequence krbtgt rotation procedure within hours
the goal is detect, contain, and recover — DCSync is the endgame attack; by the time it fires you are in incident-response mode

Prerequisites

DC audit policy: Audit Directory Service Access: Success/Failure enabled (this generates Event 4662 — not on by default on most Windows Server defaults)
SIEM parsing for Event 4662 with the Object > Properties field indexed so the replication-rights GUIDs are queryable
Authoritative list of:
Domain Controller computer accounts (the only principals that should legitimately perform DCSync)
Principals with effective replication rights via BloodHound Cypher query
Allowlist of legitimate ntds.dit access sources (backup software, AD recovery tooling)
Pre-staged krbtgt rotation script (Microsoft's New-KrbtgtKeys.ps1 or equivalent) tested in a recovery exercise — runbook ready, not improvised under pressure
Optional: honey-principal with replication rights and a strong unused password — any DCSync attempt against this principal is high-confidence adversary activity

Detection steps

This playbook pairs with DCSync and ntds.dit Extraction. Read offense + defense as the paired-reading discipline requires.

Phase 0 — Baseline (do once, refresh monthly)

Enumerate effective replication rights. In BloodHound: cypher MATCH (n)-[r:DCSync|GetChanges|GetChangesAll]->(:Domain) RETURN DISTINCT n.name AS principal, labels(n)[0] AS type Only Domain Controller computer accounts and explicit Tier 0 admins should appear. Anything else is a finding — remediate before relying on detection.
Inventory DC computer accounts. SIEM lookup table populated with the SAM accounts and source IPs of every DC. Detection rules join against this list.
Inventory legitimate ntds.dit access. Backup vendors (Veeam, Commvault, native Windows Backup), AD recovery tooling, and nothing else should ever touch the file. Document the allowlist.
Audit baseline krbtgt rotation cadence. If krbtgt was last rotated more than a year ago, schedule a single proactive rotation before relying on this playbook — the password-history window assumes the current krbtgt is the only valid key.

Phase 1 — Detect classic DCSync (DRSUAPI)

Operator signature: Event 4662 on a Domain Controller with Properties containing the replication GUIDs, where the Subject is not a DC computer account.

1. Replication-rights GUID rule (highest fidelity). Alert when Event 4662 contains any of: - 1131f6aa-9c07-11d1-f79f-00c04fc2dcd2 (DS-Replication-Get-Changes) - 1131f6ad-9c07-11d1-f79f-00c04fc2dcd2 (DS-Replication-Get-Changes-All) - 89e95b76-444d-4c62-991a-0facbeda640c (DS-Replication-Get-Changes-In-Filtered-Set) text index=wineventlog EventCode=4662 (Properties="*1131f6aa-9c07-11d1-f79f-00c04fc2dcd2*" OR Properties="*1131f6ad-9c07-11d1-f79f-00c04fc2dcd2*" OR Properties="*89e95b76-444d-4c62-991a-0facbeda640c*") | lookup dc_computer_accounts SubjectUserSid AS legitimate_dc | where isnull(legitimate_dc) Every match is an immediate high-severity alert. There are essentially zero legitimate false positives.

2. Per-account targeting profile. If multiple 4662 events for replication GUIDs fire from one source against many distinct target accounts, the operator is performing -just-dc (full dump) rather than -just-dc-user krbtgt (surgical). The full-dump variant is louder but more frequently used by less-careful operators; the surgical variant requires faster response because the goal is usually one account (krbtgt).

3. Source-host process correlation via EDR. Cross-reference the source IP from the 4662 event against EDR process telemetry on that host. Look for secretsdump.py, mimikatz.exe, Rubeus.exe, Impacket Python signatures, or unsigned binaries spawning during the alert window. A Windows host running secretsdump.py is a confirmed adversary; a Linux Impacket source is the canonical Kali/operator-host signature.

Phase 2 — Detect ntds.dit offline extraction

Operator signature: direct access to C:\Windows\NTDS\ntds.dit on a DC, or VSS shadow-copy creation followed by file access, or ntdsutil invocation with ifm / IFM subcommands.

1. Sysmon Event 11 (file create) on shadow-copy paths. text index=sysmon EventCode=11 TargetFilename="\\GLOBALROOT\Device\HarddiskVolumeShadowCopy*\Windows\NTDS\ntds.dit" Any match is a high-fidelity alert. Legitimate backup software uses VSS but does not extract individual files this way.

2. Process creation: ntdsutil, vssadmin create shadow, wbadmin start backup. text index=wineventlog EventCode=4688 (NewProcessName="*\\ntdsutil.exe" OR (NewProcessName="*\\vssadmin.exe" AND CommandLine="*create shadow*") OR (NewProcessName="*\\wbadmin.exe" AND CommandLine="*ntds*")) Alert with allowlist suppression for known-legitimate backup-software parent processes.

3. Sysmon Event 1 with ntdsutil + IFM argument pattern. ntdsutil "ac i ntds" "ifm" "create full c:\extract" q q is the canonical operator command. The argument string is recognizable; alert specifically on it.

4. EDR file-handle telemetry on ntds.dit. Modern EDR products (Defender for Endpoint, CrowdStrike, SentinelOne) emit file-handle-open events for sensitive AD database paths. Cross-reference with the legitimate-backup allowlist; everything else alerts.

Phase 3 — Detect DCShadow (rogue-DC registration)

Operator signature: anomalous changes to AD configuration container objects — a non-DC host registering itself as a DC, then walking back the registration after pushing changes.

Event 4929 / 4928 (replication source replica added/removed) outside the change-management window.
Event 5136 (directory service object modified) on objects in CN=Sites,CN=Configuration,... with SubjectUserSid that is not a DC or scheduled-task service.
Sysmon network connections to TCP 135 (RPC) + dynamic high ports from a non-DC host with a target of a DC — the DRSUAPI inbound-replication signature when the source is impersonating a DC.

DCShadow is rarer and stealthier than classic DCSync; detection often relies on Microsoft Defender for Identity's purpose-built rule rather than raw Event-Log queries.

Phase 4 — Investigation and response

Once any Phase 1–3 alert fires:

Treat as a confirmed compromise until proven otherwise. False-positive rate on these rules is near zero; the response posture should reflect that.
Identify the source host and principal. Phase 1: pull SubjectUserSid and source IP from the 4662 event. Phase 2: pull the parent process tree from EDR. Build the operator timeline for the last 24 hours from that source.
Determine extraction scope. - -just-dc-user <target> → only the named account's hash is compromised. Surgical response: rotate that account's password. - -just-dc (full dump) → every account in the domain is compromised, including krbtgt. Full-domain response. - ntds.dit + SYSTEM hive copied → same as full dump.
Initiate krbtgt rotation if krbtgt was in scope. Use Microsoft's New-KrbtgtKeys.ps1 script. Twice in sequence, with ≥ 10 hours between rotations to let replication and the password-history window roll over. A single rotation is insufficient because the previous krbtgt hash remains valid.
Disable the source account. Force re-authentication for every Tier 0 admin (assume their credential may have enabled the attack chain). Audit every authentication event from the source IP in the past 30 days.
Audit Golden / Silver Ticket use. Any TGT or TGS for krbtgt or service accounts issued between the extraction time and the rotation completion is potentially forged. Cross-reference with detect-kerberoasting-and-as-rep-roasting alerts in the same window.
Open an IR ticket with: triggering event(s), source host + principal + IP, list of extracted accounts (if known), krbtgt rotation status, Golden/Silver Ticket audit findings, full BloodHound snapshot before and after the suspected extraction.

Validation clues

High-confidence DCSync: Event 4662 with replication GUIDs from a non-DC SubjectUserSid. Single event is sufficient.
High-confidence ntds.dit extraction: Sysmon Event 11 on a shadow-copy path ending in ntds.dit, or ntdsutil invocation with IFM/ifm arguments outside the change-management window.
Honey-principal fired: any DCSync attempt against the honey replication-rights principal. Near-zero false positive — investigate as confirmed adversary activity.
Chain evidence: the same source IP/principal that triggered a Kerberos alert (detect-kerberoasting-and-as-rep-roasting) in the previous hours, now performs DCSync. Confirms a Kerberoasting → cracked-service-account → DCSync chain — the canonical AD endgame.
Forged-ticket evidence post-rotation: Event 4769 (TGS issued) where the ticket's krbtgt encryption type does not match either the current or previous krbtgt key. Indicates a Golden Ticket forged with an old, leaked krbtgt hash still in use after a single (insufficient) rotation.

Mitigation / remediation

On the defender side, these are durable controls (not per-alert fixes):

Monthly BloodHound audit of effective replication rights. Query MATCH (n)-[:DCSync|GetChanges|GetChangesAll]->(:Domain) RETURN n.name. The healthy result is "only DCs and explicit Tier 0 admins". Anything else is a finding to remediate. See DCSync §Detection-and-defense item 1.
Quarterly proactive krbtgt rotation. Twice in sequence, on a quarterly cadence, even without suspected compromise. Shortens the worst-case Golden Ticket window from "decades" to "≤ 3 months".
Tier 0 administration discipline. PAWs, Authentication Policies, Protected Users membership for all Tier 0 admins. The structural defense that prevents the credential-theft chain ending at DCSync.
Audit Exchange Trusted Subsystem and related Exchange-installed groups. Historical Exchange installations grant effective DCSync to multiple groups. Decommissioning Exchange does not automatically remove these grants — explicit audit-and-remove required.
Disable NTLMv1 domain-wide. NTLM-relay-to-DCSync chains exist; eliminating the relay surface forces operators into Kerberos-only paths that this playbook detects better.
Plant honey-principals. A fake account with DS-Replication-Get-Changes-All and a strong unused password is by definition adversary-only traffic when accessed.

What does not work as a primary defense

Blocking DRSUAPI at the network level. DCs must replicate; blocking the protocol breaks the directory.
Trusting that "only DCs have replication rights". Verify with BloodHound. Exchange-installation legacy grants routinely produce surprise principals.
Single krbtgt rotation. Password history retains the previous hash; one rotation does not invalidate leaked material.
Trusting low alert volume. A single 4662 event with a replication GUID from a non-DC source is the signature. Volumetric thresholds miss the surgical operator.
Trusting EDR alone. Modern operators use signed Microsoft binaries (Impacket via Python, in-memory secretsdump variants) that bypass binary-name signatures. The Event-Log-side detection is structural; EDR is the corroboration layer.

Logging / forensics

Retain DC Security logs for ≥ 90 days. DCSync followed by Golden Ticket use often spans weeks; retention is the floor that supports forensic timeline reconstruction.
Tag every alert with: source IP, source SubjectUserSid, target account(s), replication GUID(s) used, allowlist status (was the source a legitimate DC?), krbtgt rotation status at time of alert, correlated Kerberoasting/AS-REP alerts in the prior 30 days.
Capture the full DRSUAPI RPC traffic from the source if possible. Network telemetry (Zeek dce_rpc.log, Sysmon Event 3) from the source IP during the alert window often reveals the operator's tool fingerprint (Impacket vs Mimikatz vs custom).
Snapshot BloodHound before and after the suspected compromise. Comparing graph states reveals whether the operator added DCSync rights to other principals post-compromise (a common persistence move).
Cross-reference with attack-path correlation. A single DCSync alert is high-confidence; a DCSync alert preceded by Kerberoasting, BloodHound collection, and lateral SMB activity is a confirmed kill chain ready for incident-response handoff.

Operational safety

never auto-disable replication on a DC computer account in response to an alert — the DC must replicate to remain in the directory.
never rotate krbtgt only once after suspected compromise; always twice in sequence with replication-wait between.
always maintain a documented allowlist of legitimate ntds.dit access sources (backup software, AD recovery tooling) before relying on Phase 2 detection rules.
always test the detection rules against an authorized internal lab running DCSync before relying on them in production. Untested rules are theater.
always treat a Phase 1 alert as a confirmed compromise until proven otherwise — the false-positive rate is near zero and the cost of slow response is total-domain compromise.

DCSync and ntds.dit Extraction — the offense playbook this one mirrors.
Golden Ticket and krbtgt Compromise — the immediate-next attack after DCSync; the reason krbtgt rotation is the response priority.
Silver Ticket and Service Account Persistence — the service-account-scoped equivalent after DCSync recovers a service-account hash.
krbtgt Rotation and Tier 0 Recovery — the canonical recovery procedure after suspected krbtgt compromise.
BloodHound — the tool for both the offensive enumeration of replication rights and the defensive audit.
Tier 0 Administration — the structural defense that makes the chain to DCSync impossible.
detect-kerberoasting-and-as-rep-roasting — the upstream detection that frequently precedes a DCSync alert in the same chain.
Windows Event Logs — Event 4662 + replication GUIDs reference; Event 4929 / 5136 (DCShadow indicators).
Behavioral vs Signature Detection — the right framing for DCSync detection; no useful signature exists at the bug-class layer.
Attack Path Correlation — multi-stage correlation across recon + Kerberoast + DCSync + Golden Ticket.
run-scan-pipeline / detect-external-scan-pipeline — the parallel offense/defense playbook pair pattern this playbook follows.
Attacker-Defender Duality — the meta-principle.

References

Foundational: MITRE ATT&CK T1003.006 — OS Credential Dumping: DCSync — https://attack.mitre.org/techniques/T1003/006/
Research / Deep Dive: Sean Metcalf (ADSecurity.org) — Mimikatz DCSync Usage, Exploitation, and Detection — https://adsecurity.org/?p=1729
Foundational: Microsoft Learn — AD DS replication permissions and audit guidance — https://learn.microsoft.com/en-us/troubleshoot/windows-server/active-directory/ad-replication-event-id-1864
Recovery: Microsoft Security Blog — KRBTGT Account Password Reset Scripts — https://www.microsoft.com/en-us/security/blog/2015/02/11/krbtgt-account-password-reset-scripts-now-available-for-customers/
Research / Deep Dive: Microsoft Defender for Identity — DCSync attack detection — https://learn.microsoft.com/en-us/defender-for-identity/credential-access-alerts