[Advanced] — the blue-team half, and why understanding the attacks is what makes the defence make sense.

The kid version first

Every attack in this folder abuses a default, a misconfiguration, or a credential left where an attacker could reach it. So defence is the mirror image: turn off insecure defaults, remove the over-generous permissions, keep powerful credentials off ordinary machines, and watch the graph for the paths attackers walk.

You cannot defend AD well without knowing how it’s attacked — which is exactly why this note comes last.

The single most important idea: tiering

Powerful credentials must never touch machines an attacker can compromise.

The whole engine of AD compromise is: attacker owns a workstation → a Domain Admin logged into it → mimikatz steals their credentials → domain over. Break that chain and most attacks stall.

Microsoft’s tiered administration model:

Tier 0   Domain Controllers, AD, KRBTGT — the crown jewels
Tier 1   Servers and applications
Tier 2   Workstations

The rule: a Tier 0 admin credential is used ONLY on Tier 0 systems. A Domain Admin never logs into a workstation or a regular server — so even if that machine is owned, there’s no DA credential in its memory to steal.

This is the highest-impact control by far, and the hardest to implement because it changes how admins work. Most breached organisations had Domain Admins logging into everything.

Countering each attack

Mapping defences to the attacks, because that’s how they make sense:

AttackDefence
KerberoastingManaged Service Accounts (gMSA) — 120-char auto-rotated passwords, uncrackable. Where you must use a human service account, give it a long random password
AS-REP roastingDon’t disable Kerberos pre-auth. Audit for accounts that have it disabled
Pass-the-hashLAPS — unique random local admin password per machine, so one hash owns one box, not the fleet. Tiering. Disable NTLM where possible
mimikatz / LSASSCredential Guard (virtualisation-isolated LSASS), EDR, disable WDigest, tiering so there’s nothing worth stealing
DCSyncAudit who has replication rights — it should be only DCs and DAs. Alert on replication from a non-DC
Golden TicketRotate KRBTGT periodically (twice, correctly). Monitor for anomalous TGT lifetimes
ACL abuseAudit ACLs (BloodHound defensively), remove accumulated over-delegation, protect privileged groups with AdminSDHolder
Delegation abuseMark Tier 0 accounts “sensitive, cannot be delegated.” Audit unconstrained delegation — there should be almost none. Patch the coercion bugs (PetitPotam)
ADCS (ESC1–16)Audit templates with certipy; fix enrollment permissions and SAN settings
Responder / NTLM relayDisable LLMNR/NBT-NS. Enforce SMB signing. Enforce LDAP signing/channel binding

Notice the pattern: run the attacker’s own tools against yourself. BloodHound, certipy, nxc and PingCastle are blue-team tools too — they find the paths before an attacker does.

Detection — because prevention is never complete

Enumeration and lateral movement use legitimate protocols (03), so detection is behavioural, not signature-based → security operations:

  • The right event IDs — 4768/4769 (Kerberos tickets, for roasting), 4662 (DCSync-shaped replication), 4728/4732 (privileged group changes), 4624 with anomalous logon patterns
  • Honeytokens — a fake service account with an SPN and a tempting name. Nobody legitimate ever requests its ticket, so a 4769 for it is a near-certain Kerberoast alert. Cheap, high-signal, and one of the best AD detections available
  • Behavioural baselines — one account authenticating to twenty machines, a workstation querying every ACL, a TGT with a ten-year lifetime. Microsoft Defender for Identity and similar watch for exactly these patterns
  • The BloodHound query, run by defenders — “what’s the shortest path to DA?” is a remediation backlog, not just an attack plan

The hard truth about recovery

Assume breach. A determined attacker with a foothold and time will usually find a path in a large AD environment, because the misconfigurations accumulate over decades. So the goal isn’t a perfect wall — it’s:

  1. Raise the cost — tiering, gMSA, LAPS, removed delegation make the easy paths disappear
  2. Detect fast — behavioural monitoring and honeytokens catch the attacker mid-chain
  3. Recover credibly — and know that KRBTGT compromise is genuinely hard to recover from (double rotation, and forest-wide trust considerations), which is why keeping attackers away from Tier 0 is worth so much

Where cloud changes the picture

Azure AD / Entra ID is a different model (OAuth, conditional access, no Kerberos), and hybrid environments — on-prem AD synced to Entra — introduce their own attack paths (Azure AD Connect compromise, PRT theft, illicit consent grants). Hybrid is now the common case, and it means an on-prem compromise can pivot to the cloud tenant and vice versa → cloud security.

Key insight

Every AD defence is the mirror of a specific attack, which is why you can’t defend it without understanding the offence — tiering kills credential theft, gMSA kills Kerberoasting, LAPS kills hash reuse, honeytokens catch roasting for free. The defining reality is that prevention is never complete in a decades-old domain, so the winning strategy is: make the easy paths vanish, watch behaviourally for the attacker walking the hard ones, and keep the KRBTGT-grade credentials on Tier 0 machines an attacker can’t reach.

Source: [reference] — Aug 2026.

↑ Back to Active Directory