Detect Kerberoasting and AS-REP Roasting

Goal

Detect both Active Directory Kerberos credential-attack families (Kerberoasting and AS-REP Roasting) via behavioral analysis on Event IDs 4768/4769 and complementary telemetry, with high enough fidelity to identify the source account, attacker host, and scope within minutes — and durable enough to survive small operator tradecraft variations (encryption type, request rate, account selection).

Assumptions

you collect Windows Security event logs from all Domain Controllers into a SIEM (Splunk, Sentinel, Elastic, or equivalent) with ≥ 30 days retention
you have an authoritative list of service accounts (accounts with servicePrincipalName attribute set) and an audit of accounts with DONT_REQ_PREAUTH set
you can hot-disable a compromised account and require Tier 0 approval for re-enable
the goal is detect, contain, investigate — not prevention; the protocol features these attacks exploit are not patchable as bugs

Prerequisites

DC audit policy: Audit Kerberos Authentication Service: Success/Failure and Audit Kerberos Service Ticket Operations: Success/Failure both enabled (these generate 4768 and 4769 respectively) — see Windows Event Logs for the full audit-policy + Event-ID reference
SIEM tables for Event 4768 and 4769 with the field for TicketEncryptionType (4769) or PreAuthType (4768) parsed and indexed
A baseline of normal Kerberos behavior per service account: typical request rate, typical source IPs, typical encryption types
An authoritative list of accounts that legitimately have DONT_REQ_PREAUTH set (should be ≤ 1 in a healthy environment, ideally zero)
Optional: honey accounts with SPN set and/or pre-auth disabled (asymmetric high-confidence detection)

Detection steps

This playbook pairs to two offense notes: Kerberoasting and AS-REP Roasting. Read offense + defense as the paired-reading discipline requires.

Phase 0 — Baseline (do once, refresh quarterly)

Inventory accounts with SPNs. These are your Kerberoasting targets. Audit which are legitimate service accounts vs which are misconfigurations. text Get-ADUser -Filter {ServicePrincipalName -like "*"} -Properties ServicePrincipalName
Inventory accounts with DONT_REQ_PREAUTH. These are your AS-REP roasting targets. The healthy count is zero. text Get-ADUser -Filter 'DoesNotRequirePreAuth -eq $true' -Properties DoesNotRequirePreAuth
Baseline encryption types per account. Run for 7 days and record the dominant TicketEncryptionType per service account (should be AES256 or AES128 — value 0x12 or 0x11 — in any modern environment).
Baseline request rates per service account. A real service account typically requests a small number of TGS tickets per day, from a small set of source hosts. Record per-account.
Audit and remediate Phase 0 findings before relying on detection. Reducing the surface area is more valuable than detecting attacks against unnecessary surface area.

Phase 1 — Detect Kerberoasting

Operator signature: bulk TGS-REQ traffic from one source against many service accounts, often with RC4-HMAC encryption type, often within seconds.

High-volume TGS request rule. Alert when a single account requests TGS for ≥ N distinct service accounts within 60 s. N depends on baseline; 5–10 is a strong starting point. text # Splunk-ish pseudocode index=wineventlog EventCode=4769 TicketEncryptionType IN ("0x17","0x12","0x11") | bin _time span=60s | stats dc(ServiceName) AS uniq_services BY _time, TargetUserName, IpAddress | where uniq_services >= 5
RC4-HMAC anomaly rule. Alert on TGS-REQ with TicketEncryptionType=0x17 (RC4-HMAC) for any service account whose baseline is AES. The attacker frequently downgrades to RC4 to make offline cracking economically viable. text index=wineventlog EventCode=4769 TicketEncryptionType="0x17" | lookup service_account_baseline ServiceName OUTPUT baseline_enctype | where baseline_enctype != "0x17"
Service-account behavioral deviation rule. Alert when a service account's TGS request source deviates from baseline. Real service accounts are used from the same small set of hosts; the attacker host is new.
Per-source LDAP+Kerberos correlation. A Kerberoasting operator usually queries LDAP for service accounts ((&(objectClass=user)(servicePrincipalName=*))) immediately before issuing TGS-REQs. Alert on LDAP-with-SPN-filter followed within 60 s by anomalous TGS-REQ from the same source.

Phase 2 — Detect AS-REP Roasting

Operator signature: AS-REQ from one source for one or more accounts, with Pre-Authentication Type: 0 (no PA-ENC-TIMESTAMP provided).

PreAuthType=0 rule (highest fidelity). Event 4768 with PreAuthType=0 is by definition pre-authentication-disabled traffic. In an environment where no account should have DONT_REQ_PREAUTH set, any event of this type is an immediate high-severity alert. text index=wineventlog EventCode=4768 PreAuthType="0" | lookup preauth_disabled_allowlist TargetUserName OUTPUT allowed | where isnull(allowed)
AS-REQ source diversity rule. A single source sending AS-REQ against many usernames (especially against non-existent ones, which generate KRB5KDC_ERR_C_PRINCIPAL_UNKNOWN) is a username-enumeration + roasting signature. Alert on the combination.
AS-REQ + KRB-ERROR pattern. A roasting operator often gets a mix of KRB-AS-REP (success against pre-auth-disabled accounts) and KRB5KDC_ERR_PREAUTH_REQUIRED (the expected response for protected accounts) from one source. The pattern is distinctive.

Phase 3 — Investigation and response

Once a Phase 1 or Phase 2 alert fires:

Identify the source host. Pull IpAddress (4769) or Source Network Address (4768) from the triggering event. Correlate with DHCP and AD computer logs to identify the host.
Pull all Kerberos events from that source in the last 24 hours. Build the operator's activity timeline.
Hash the timeline. If the source also performed LDAP enumeration, SMB session enumeration, or BloodHound-style ACL queries within the same window, you are looking at an authenticated foothold doing AD reconnaissance (BloodHound).
Determine compromise scope. - For Kerberoasting: every account whose TGS was requested in the timeline should be considered "hash potentially exfiltrated, password potentially crackable" until proven otherwise. Rotate every such account's password. - For AS-REP roasting: the named target accounts should be rotated and the DONT_REQ_PREAUTH flag cleared.
Disable the source account. Until you can prove the source is legitimate or remediate it, treat it as hostile.
Open IR ticket with: triggering event(s), source host/IP/user, list of affected service accounts, rotation status of each, BloodHound output snapshot.

Validation clues

High-confidence Kerberoasting: one source account requests TGS for ≥ 5 distinct service accounts within 60s, with RC4-HMAC encryption type against an AES-baseline environment.
High-confidence AS-REP roasting: Event 4768 with PreAuthType=0 against any account not on the explicit allowlist.
Honey account fired: TGS-REQ or AS-REQ against the honey SPN/account. Near-zero false-positive — investigate immediately as confirmed adversary activity.
Behavioral cluster: the same source IP that triggered the Kerberos alert also performed LDAP enumeration with SPN filter or BloodHound-style ACL queries in the same window. Confirms recon-phase activity.
Cracking evidence: if the operator successfully cracked a service account password, you may see a 4768 (TGT request) success from a different source IP using the same account's credential within hours. Detection-ready.

Mitigation / remediation

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

Migrate all service accounts to Group Managed Service Accounts (gMSA). gMSA passwords are 240-byte random values rotated by AD on a schedule, making offline cracking infeasible regardless of encryption type. See gMSA and modern service account hardening for the migration discipline.
Enforce AES-only encryption types domain-wide (KerberosEncryptionTypes GPO set to AES128/AES256 only). Removes the RC4 cracking advantage. Even a weak password becomes uncrackable in practical timeframes against AES-PBKDF2.
Remove DONT_REQ_PREAUTH from every account. The bitmask audit query ((userAccountControl:1.2.840.113556.1.4.803:=4194304)) returns the list. Re-enable pre-auth or disable the account. The healthy long-term state is zero pre-auth-disabled accounts.
Remove privileged service accounts from Tier 0 groups. Service accounts should not be in Domain Admins. Period. See Tier 0 Administration for the structural framing.
Honey accounts. Plant a fake service account with SPN set and another with DONT_REQ_PREAUTH. Both with strong random passwords nobody uses. Any TGS-REQ or AS-REQ for either is, by definition, adversary activity.
Disable NTLMv1 domain-wide. AS-REP and TGS hashes can sometimes be downgraded to NTLMv1 in cross-trust scenarios. Block NTLMv1 outright.

What does not work as a primary defense

Blocking outbound Kerberos. Kerberos is the authentication protocol; you cannot block it without breaking AD.
Renaming service accounts to obscure names. LDAP returns the canonical attribute. Security by obscurity.
Trusting that "we audited a year ago". userAccountControl flags drift; new service accounts get created with weak passwords; new Exchange installations grant unexpected rights. The audit cadence must be quarterly.
Disabling Mimikatz / Impacket binaries. The protocol is the attack surface, not the tool. Block-listing tools is a delaying tactic, not a defense.
Trusting low alert volume. A single TGS-REQ for krbtgt or a single PreAuthType=0 event is the signature. Volumetric thresholds miss the surgical operator.

Logging / forensics

Retain DC Security logs for ≥ 90 days. Roasting → cracking → reuse chains often span days or weeks. Without retention, the chain is invisible.
Tag every alert with: source account, source IP, target service accounts, encryption type, allowlist status (was this an expected pre-auth-disabled account?). These tags drive trend analysis.
Capture the full TGS-REQ / TGS-REP exchange where possible. Wireshark on a Domain Controller during an active alert gives you the operator's exact request payload — useful for tool fingerprinting and forensics.
Correlate with attack-path correlation across the chain. A single Kerberoast event is suspicious; a Kerberoast event followed by a cracked credential authenticating from a different source IP days later is a confirmed kill chain.

Operational safety

never disable a service account "to be safe" without coordinating with the application team. Service accounts in production runners can cause outages.
never rely on a single detection rule. Phase 1 alert + LDAP-enumeration correlation + behavioral deviation produces much higher-fidelity findings than any single rule alone.
always maintain a written allowlist of accounts with legitimate pre-auth-disabled or weak-encryption-type configurations. Without it, every alert is potentially a false positive.
always rotate every Kerberoasted service account's password as part of remediation, even if you believe the operator could not crack it. The hash was exfiltrated; the only durable response is rotation.
always test detection rules against an authorized internal lab running the offense attacks before relying on them. Untested rules are theater.

Kerberoasting — the offense playbook half of Phase 1.
AS-REP Roasting — the offense playbook half of Phase 2.
BloodHound — the recon tool the operator usually pairs with these attacks; alert correlation finds both.
DCSync — the endgame after a cracked service-account password yields DCSync rights.
gMSA hardening — the structural mitigation referenced repeatedly above.
Tier 0 Administration — the structural defense that makes the post-Kerberoast escalation chain impossible.
Behavioral vs Signature Detection — the right framing for AD attack telemetry.
Attack Path Correlation — multi-stage correlation across recon + Kerberoast + cracked-credential reuse.
run-scan-pipeline and detect-external-scan-pipeline — the parallel offense/defense pair pattern this playbook follows.
Attacker-Defender Duality — the meta-principle.

References

Foundational: MITRE ATT&CK T1558.003 — Kerberoasting — https://attack.mitre.org/techniques/T1558/003/
Foundational: MITRE ATT&CK T1558.004 — AS-REP Roasting — https://attack.mitre.org/techniques/T1558/004/
Research / Deep Dive: Sean Metcalf (ADSecurity.org) — Detecting Kerberoasting Activity — https://adsecurity.org/?p=3458
Research / Deep Dive: Microsoft Threat Intelligence — How attacks abuse Kerberos: detection and mitigations — https://www.microsoft.com/en-us/security/blog/2022/01/26/evolving-kerberos-attack-detection/