LLMNR Poisoning Explained: How the Attack Works and How to Stop It

    LLMNR poisoning turns a failed name lookup into a stolen password hash. Here's how the attack works, how NetNTLMv2 hashes get cracked or relayed, what defenders can watch, and how to shut it down.

    By Hacker Blueprint··13 min read

    The short answer

    LLMNR poisoning is an Active Directory attack that turns a failed name lookup into a stolen password hash. When a Windows machine can't resolve a hostname through DNS, it shouts a query to the whole local network using LLMNR or NBT-NS. Any machine on that network can answer "that's me," and the victim then tries to authenticate to the attacker. The attacker captures the username and a NetNTLMv2 hash, then cracks it offline or relays it to another machine.

    The fix is to stop the broadcasts. Turn off multicast name resolution with a Group Policy setting, disable NetBIOS over TCP/IP, and require SMB signing so captured hashes can't be relayed. MITRE ATT&CK tracks the technique as T1557.001. This guide is for OSCP® students learning Active Directory (AD) and for defenders who want to know what to watch and shut off.

    We're Hacker Blueprint. Our instructor holds the OSCP, our academy has more than 600 students, and around 7,000 people have claimed our prep materials.

    Our video "Responder for NTLM Hashes (Full Guide for OSCP, CPTS, CRTO & More!)" shows the capture in a lab: an intro to hash stealing at 0:00, setting up the attack at 1:12, the capture at 4:03, cracking the result at 6:17, and where it fits in a methodology at 6:43. It runs about nine minutes.

    What is LLMNR and why does it exist?

    LLMNR stands for Link-Local Multicast Name Resolution. It's a fallback Windows uses when DNS can't answer. The computer sends a query to every host on the local link and asks, "does anyone here know this name?" NBT-NS (NetBIOS Name Service) is the older version of the same idea, and mDNS (Multicast DNS) is the newer cross-platform one. All three resolve names on the local network when normal DNS comes up empty.

    LLMNR is defined in RFC 4795, published in January 2007 as an Informational document rather than an internet standard. It listens on UDP port 5355 and uses the multicast addresses 224.0.0.252 (IPv4) and FF02::1:3 (IPv6). NBT-NS uses UDP 137, and mDNS uses UDP 5353. The RFC itself says this was never meant to be a main resolver: "use of LLMNR as a primary name resolution mechanism is NOT RECOMMENDED." It's a convenience for small networks and a safety net when DNS is missing. That convenience is the whole problem, because these protocols trust whoever answers first.

    How does LLMNR poisoning work?

    The attack needs no exploit and no password. It only needs a position on the same local network as the victim.

    A failed lookup happens constantly on a busy Windows network. Someone mistypes a share name, an old login script points at a server that no longer exists, or a machine reaches for a printer that has moved. DNS returns nothing, so Windows falls back to LLMNR or NBT-NS and broadcasts the name to everyone on the link. The host it's looking for usually doesn't exist, which is why DNS failed in the first place. The attacker's machine answers instead and claims to be that host.

    The victim believes the answer and connects. If the service is SMB, the victim's machine tries to authenticate automatically, and that handshake carries the username and a NetNTLMv2 response derived from the user's password. MITRE describes the result plainly: "the username and NTLMv2 hash will then be sent to the adversary controlled system."

    RFC 4795 predicted this in its security section. Because "multicast LLMNR queries are propagated to all hosts on the link," an on-link attacker can "spoof LLMNR responses without having to guess the value of the ID field in the query." The attacker only has to beat the legitimate reply, and when the name doesn't resolve at all, "spoofing a response in a timely way is not difficult." The design assumes everyone on the network is trustworthy, and a corporate LAN rarely is.

    What is the Responder tool?

    Responder is the tool named in almost every writeup. Its own description is blunt: "Responder is a LLMNR, NBT-NS and MDNS poisoner." It was written by Laurent Gaffié, and the commonly used version is the fork on the lgandx GitHub account. When it runs on a network, it listens for broadcast name lookups, answers them, and runs rogue authentication servers that collect the hashes victims send.

    Responder also has an Analyze mode that watches name-resolution requests passively without answering any. That mode cuts both ways. For a tester, it shows which hosts are noisy without tipping anyone off. For a defender, the same passive view measures how much LLMNR and NBT-NS traffic a network leaks before the protocols get turned off.

    NetNTLMv2 hashes: cracking vs relaying

    A captured NetNTLMv2 hash is a challenge-response value, not the stored password hash. You can't pass it the way you'd pass an NTLM hash. There are two ways to turn it into access.

    The first is cracking. You feed the hash to a cracker and guess passwords offline until one produces the same response. hashcat lists NetNTLMv2 as mode 5600 and the older NetNTLMv1 as mode 5500. The domain never sees the guesses, so account lockout never triggers. A weak or reused password falls fast. A long, random one may never fall, which is why password strength is a defense here.

    The second is relaying. Instead of cracking, the attacker forwards the whole authentication to a third machine and logs in as the victim there. MITRE notes that "the NTLMv1/v2 hashes can be intercepted and relayed to access and execute code against a target system." Relaying skips cracking entirely, so even a 30-character password doesn't save you. The thing that stops a relay is SMB signing, covered under prevention.

    For the Kerberos attacks that work from a similar early position, our AS-REP roasting guide and Kerberoasting attack guide go into the same depth.

    Where LLMNR poisoning fits in an OSCP-style AD methodology

    This is an internal-network technique. It does nothing from the outside. You need a foothold on the local network first, whether that's a compromised machine, a drop box, or a seat on the client LAN. Once you're on the link, poisoning is often the fastest route from "no credentials" to "a valid domain user," and that one user unlocks Kerberoasting, BloodHound collection and the rest of the chain.

    The OSCP exam adds a rule. OffSec allows Responder as a tool but bans poisoning and spoofing, so you can't actively answer other machines' lookups on the exam network. Read OffSec's exact wording in its exam FAQ before you plan to use Responder there. The mechanism still matters for the exam, because it explains why NetNTLMv2 hashes turn up and what they're worth. Our OSCP Active Directory methodology explains that line, and our guide to OSCP allowed tools has the full tool rules.

    To see these attacks worked on screen, our Hacking Active Directory playlist covers AD methodology from the basics to full guides, with BloodHound, AS-REP roasting, Kerberoasting, certificate attacks and lateral movement.

    How to detect LLMNR and NBT-NS poisoning

    MITRE's detection strategy DET0462, created October 21, 2025 and last modified May 12, 2026, lists what to watch:

    • Unexpected responders. "Unusual responses to LLMNR (UDP 5355) or NBT-NS (UDP 137) queries from unauthorized hosts." On a healthy network, almost nothing should answer these broadcasts, so one host that suddenly does stands out.
    • SMB auth after a poisoned lookup. "Abnormal SMB authentication attempts correlated with poisoned LLMNR/NBT-NS sessions." A name lookup followed instantly by an SMB login to an odd host is the fingerprint.
    • Registry changes that re-enable the protocols. Modifications to HKLM\Software\Policies\Microsoft\Windows NT\DNSClient\EnableMulticast, plus suspicious service creation (Windows Security event 4697).

    Honey lookups are a strong trick: request a hostname that doesn't exist and shouldn't ever be queried, then alert if anything answers. A decoy account does the same for the login side. Microsoft Defender for Identity raises a "Honeytoken authentication activity" alert when a decoy account that should sit unused is suddenly used, and it flags relay directly with "Suspected NTLM relay attack (Exchange account)," external ID 2037. On endpoints, Sysmon event ID 3 records network connections, event ID 13 records registry value changes, and event ID 22 records DNS queries, all useful to hunt in a SIEM.

    How to disable LLMNR and prevent poisoning

    The fix is to remove the broadcasts and block the relay. MITRE's mitigations line up as disable the feature (M1042), filter the traffic (M1037), and segment the network (M1030). In practice:

    1. Turn off LLMNR with Group Policy. Enable "Turn off multicast name resolution" under Computer Configuration > Administrative Templates > Network > DNS Client. It sets EnableMulticast to 0 at HKLM\Software\Policies\Microsoft\Windows NT\DNSClient, which kills LLMNR domain-wide.
    2. Disable NetBIOS over TCP/IP. LLMNR isn't the only broadcast protocol, so handle NBT-NS separately. Microsoft's DHCP method uses the "001 Microsoft Disable Netbios Option" scope option set to 0x2, or you can set it per adapter on the WINS tab. Disable mDNS too if nothing needs it.
    3. Require SMB signing. Signing stops the relay path. Microsoft says it helps "protect your deployments from relay and spoofing attacks," and MITRE notes that "Enabling SMB Signing can stop NTLMv2 relay attacks." On Windows 11 version 24H2 Enterprise, Pro and Education it's now required by default in both directions, and Windows Server 2025 requires outbound signing by default.
    4. Enforce strong passwords. A hash that can't be relayed can still be cracked offline, so long, random passwords move a NetNTLMv2 hash out of a cracker's reach.
    5. Segment and monitor. Limit free authentication across the network and run the detections above so a poisoner gets caught fast.

    A ready AD lab is the best way to see both halves. The free Hacking Active Directory Course for Beginners and our free OSCP labs give you somewhere to practice safely.

    Is LLMNR poisoning still a thing in 2026?

    Yes, because the broadcast protocols are still on by default. Microsoft has said for years that it wants to move to mDNS and retire the rest. Its 2022 networking blog post "Aligning on mDNS, ramping down NetBIOS name resolution and LLMNR" calls mDNS "the modern standard for multicast name discovery" and sets the end goal that "mDNS is the only multicast name resolution protocol on by default." The same post admits the slow pace: "The default LLMNR behavior has not been changed in Windows yet."

    The wider move away from NTLM is also underway but unfinished. Microsoft's January 2026 post "Advancing Windows security: Disabling NTLM by default" plans to block network NTLM by default in "the next major Windows Server release and associated Windows client releases," with no date given. NTLM auditing is the only phase available today, on Windows Server 2025 and Windows 11 version 24H2. Until NTLM is actually disabled and these protocols are off by default, the attack keeps working, and turning them off yourself is the fix.

    FAQ

    What is LLMNR poisoning?

    LLMNR poisoning is an attack where someone on your local network answers a Windows name lookup that DNS failed to resolve, pretends to be the host the victim wanted, and captures the victim's username and NetNTLMv2 hash when the machine authenticates. The attacker then cracks the hash offline or relays it to another system. MITRE ATT&CK tracks it as T1557.001, and it usually covers NBT-NS and mDNS poisoning too, since all three protocols behave the same way.

    What is the difference between LLMNR and NBT-NS?

    They do the same job with different ages. NBT-NS (NetBIOS Name Service) is the older method on UDP 137, and LLMNR (Link-Local Multicast Name Resolution) is the newer one on UDP 5355. mDNS (UDP 5353) is the modern cross-platform version. Windows tries DNS first, then falls back to these broadcasts, so an attacker who can answer any of them can poison the lookup.

    Do you need credentials for LLMNR poisoning?

    No. You don't need a password or a prior login. You only need to be on the same local network segment as your targets, which makes this an internal-network attack rather than something you run from the internet. The hashes you capture are what give you your first set of valid credentials.

    Can you pass a captured NetNTLMv2 hash?

    No. A NetNTLMv2 hash is a challenge-response value, so it can't be used in a pass-the-hash attack the way a stored NTLM hash can. You either crack it offline to recover the password (hashcat mode 5600) or relay the authentication to another machine. Relaying is blocked by SMB signing, and cracking is blocked by a long, random password.

    How do you detect LLMNR poisoning?

    Watch for a host that answers LLMNR (UDP 5355) or NBT-NS (UDP 137) queries when it shouldn't, and for SMB logins that follow those answers to unusual destinations. MITRE's DET0462 lists both, plus changes to the EnableMulticast registry value and suspicious service creation. Honey lookups and decoy accounts catch poisoners quietly, and Defender for Identity raises a honeytoken alert when a decoy account is used.

    How do you disable LLMNR?

    Enable the "Turn off multicast name resolution" Group Policy under Computer Configuration > Administrative Templates > Network > DNS Client, which sets EnableMulticast to 0. That handles LLMNR, but you also need to disable NetBIOS over TCP/IP separately, for example through the DHCP "001 Microsoft Disable Netbios Option" set to 0x2 or the WINS tab on each adapter. Disable mDNS too if nothing in your environment relies on it.

    Is LLMNR poisoning on the OSCP?

    OffSec doesn't publish a list of exam techniques, and the PEN-200 syllabus lists "attacking Active Directory authentication" and "password attacks" without naming specific attacks. Responder is on the allowed-tools list, but poisoning and spoofing are banned on the exam, so answering other machines' lookups is off the table. Read OffSec's exact wording in its exam FAQ before you plan to use it. It's still core knowledge for OSCP-style AD practice, so learn the mechanism alongside AS-REP roasting and Kerberoasting.

    Practice on a lab you're allowed to attack

    Only run this on systems you own or have written permission to test, and keep any lab on an isolated network. The cleanest way to understand both sides is to build your own AD home lab: watch a client make a failed lookup, see what a passive capture shows, then turn on the Group Policy and NetBIOS settings above and confirm the broadcasts stop.

    For more AD practice, our free Chain_01 and Chain_03 are three-machine Active Directory labs (two Windows clients and a domain controller) that you import into VirtualBox and attack offline. You'll need 8 GB of RAM at minimum, 16 GB recommended, on an Intel or AMD machine. Get them from our free Active Directory labs.

    We also have a newer short video, "How to Steal NTLM Hashes on the OSCP (Responder)," on where NetNTLMv2 capture comes up across standalone machines, databases and web apps. You can watch it at youtube.com/watch?v=-vgekToFQRc.

    One of our students, Palak, wrote: "The AD Chains were incredibly helpful. Having a clear, repeatable methodology gave me a framework I could rely on instead of feeling completely lost whenever I got stuck. Thanks to this guidance, I went from scoring 10 points on my first attempt to passing with 70 points in under 8 hours, just three months later." (More reviews)

    Try our 30+ hour course for OSCP prep with a free trial, or start with one of our free Active Directory chains.

    Got a NetNTLMv2 hash that won't crack? Ask in the Hacker Blueprint Discord.

    Sources

    Keep reading

    Want new labs and discount codes first?

    We'll email you when a new lab drops or a sale goes live.

    Your email is safe with us. We don't spam.

    © 2026 Hacker Blueprint. All rights reserved.

    Built for hackers, by hackers.

    OSCP® and OSCP+™ are trademarks of OffSec. Hacker Blueprint is an independent training provider and is not affiliated with, endorsed by, or sponsored by OffSec. All content is our own original material and contains no OffSec course, lab, or exam content. Individual results vary: passing any exam depends on your own study and practice. * Lifetime access and lifetime updates run for as long as Hacker Blueprint operates, with no renewal and no extra charge.