Skip to content

Responder: LLMNR/NBT-NS Poisoning Explained

Responder explained: how it answers LLMNR, NBT-NS and mDNS lookups to capture Windows credentials, what a finding means, and how to shut it down.

Mark KissBy , Founder & CEOUpdated 10 min read

Active Directory Security AssessmentFrom $6,000Fixed scope, free retest, no hourly billing.

Get my fixed price

Responder is a tool that captures Windows credentials by exploiting a helpful little behavior built into every default Windows network, and it is one of the first things a tester runs once they have a network cable or a Wi-Fi connection. It requires no credentials to start, no software on any target, and no exploit in the classic sense. It simply listens for Windows machines asking for help and answers “yes, that’s me.” That is why an internal test so often begins here, and why the fix is one of the highest-value hardening steps a Windows environment can take.

The behavior it abuses: LLMNR and NBT-NS name resolution

When a Windows machine tries to reach a name that DNS cannot resolve (a mistyped share, a decommissioned server, an old printer), it does not give up. It falls back to two legacy protocols, LLMNR, which asks by multicast, and NBT-NS, which asks by broadcast, effectively shouting to the whole local network: “does anyone know where SERVER-X is?”

These protocols have no authentication. Any machine on the network can answer. Responder is the machine that always answers: “yes, that’s me, send your credentials.” The victim, believing it has found the resource it wanted, dutifully attempts to authenticate, handing over the user’s username and a hashed form of their password.

The whole attack rests on trusting an unauthenticated answer from the local network, which is exactly the assumption these protocols were built on and which no longer holds. Neither is new: the NetBIOS name service was standardized in RFC 1001 in March 1987, and LLMNR was published as RFC 4795 in January 2007.

The project’s README describes Responder as an LLMNR, NBT-NS and mDNS poisoner with built-in rogue authentication servers, so multicast DNS is the third fallback to check, not only the two legacy Windows protocols. It is still actively maintained: as of November 2026 the version string in the Responder repository reads 3.2.2.0, with commits as recent as June 2026. It also has a listen-only analysis mode that shows which of these requests your network sends without answering any of them, a useful way to size the problem before anything is captured.

Protocol Port and transport What Responder does Fix
LLMNR UDP 5355, multicast Answers the failed name lookup Group Policy: Turn off multicast name resolution
NBT-NS UDP 137, broadcast Answers the NetBIOS name query Disable NetBIOS over TCP/IP on each adapter, by DHCP option or by policy
mDNS UDP 5353, multicast Answers the multicast DNS lookup Disable or block it where clients do not need it

MITRE ATT&CK tracks the technique as T1557.001, name resolution poisoning and SMB relay.

What actually happens in the attack: how Responder works, step by step

  1. Responder listens quietly on the local network for LLMNR and NBT-NS requests. This step is passive: no scanning, no noise, nothing that looks like an attack.
  2. A Windows machine fails a name lookup (which happens constantly on real networks from typos, stale shortcuts, and old references) and broadcasts for help.
  3. Responder answers, claiming to be the requested resource.
  4. The victim sends an authentication attempt, and Responder captures the username and the Net-NTLMv2 hash of the password.
  5. That hash is then either cracked offline to recover the plaintext password, or relayed to another system (using a tool like Impacket’s ntlmrelayx) to authenticate as the victim without cracking anything.

From plugging in to holding valid credentials can take minutes, without scanning or exploiting anything.

What a finding means for you: Responder in a penetration test report

If Responder appears in a report and captured anything, it tells you two things directly (for the rest of the document, see how to read a penetration test report):

  • LLMNR and NBT-NS are enabled on your network: the legacy name-resolution fallback is active and being abused. This is the root cause, and it is a configuration, not a bug.
  • Password quality and NTLM exposure are a risk: if captured hashes cracked quickly, passwords are weak; if they were relayed successfully, SMB signing was not enforced.

Captured domain credentials are rarely the end of the story. They are the foothold. A tester feeds them into tools like NetExec to find everywhere that account works, and into BloodHound to map what it can reach in Active Directory, turning one captured credential into network-wide movement. That chain (plug in, poison, capture, crack or relay, move) is one of the most reliable paths from “physical or Wi-Fi access” to “domain compromise,” which is why it is tested so consistently. It is also why physical and wireless tests check what a network connection in your office leads to.

How defenders shut it down: disable LLMNR, NBT-NS and mDNS

This is one of the satisfying findings, because the primary fix is decisive and free:

  1. Disable LLMNR and NBT-NS. With modern DNS, these legacy fallbacks are rarely needed. Turning them off (via Group Policy for LLMNR and network configuration for NBT-NS) removes the behavior Responder depends on. The attack simply stops working. This is the single highest-value action. Review multicast DNS (mDNS) at the same time, since it is the third fallback the tool answers.
  2. Enforce SMB signing. This defeats the relay variant even where a hash is captured.
  3. Strong passwords, so any hash that is captured resists offline cracking.
  4. Network segmentation, limiting how far a poisoning attack on one segment can reach.
  5. Monitor for it: unusual name-resolution responses and authentication patterns can be detected, though prevention is far better here.

The reason testers check for this on nearly every internal engagement is that the fix is so effective and so often left undone. Disabling two legacy protocols closes one of the most common paths to compromise in Windows networks. Our Active Directory hardening checklist puts these fixes alongside the other relay, poisoning and coercion checks, each with a way to confirm it holds.

How to disable LLMNR, NBT-NS and mDNS: the exact settings

  • LLMNR. Enable the Group Policy setting Turn off multicast name resolution, under Computer Configuration, Administrative Templates, Network, DNS Client. Microsoft’s DNS client policy reference documents it; it writes the EnableMulticast value under Software\Policies\Microsoft\Windows NT\DNSClient.
  • NBT-NS. There are three routes. Per adapter: IPv4 properties, Advanced, WINS tab, Disable NetBIOS over TCP/IP. For DHCP clients: set the Microsoft vendor option 001 Microsoft Disable Netbios Option to 0x2 on the scope, as Microsoft describes in Disable NetBIOS over TCP/IP by using DHCP. On current Windows builds, the Configure NetBIOS settings policy in the same DNS Client folder can turn NetBIOS name resolution off.
  • mDNS. Where clients do not need it, block UDP 5353 at the host firewall or disable it by policy. Test first: some printer and screen-casting discovery depends on it.

Roll the change out to a pilot group, watch for applications that relied on short names, then apply it everywhere, including servers. Two related paths are not closed by these settings: proxy auto-discovery (WPAD) lookups and IPv6 DHCPv6 spoofing, the technique behind the mitm6 tool. Both need their own fix, so ask your tester whether they were checked.

How to detect LLMNR and NBT-NS poisoning

Prevention is the goal, but detection matters while the rollout is in progress and on segments you cannot change yet.

  • Watch who answers. On a network with working DNS, almost nothing should answer LLMNR on UDP 5355 or NBT-NS on UDP 137. One host answering lookups for many different names, especially names that do not exist, is a poisoner. ATT&CK’s detection guidance for T1557.001 is built on anomalous traffic on these two ports combined with SMB relay attempts.
  • Send a honeytoken lookup. From a monitoring host, request a random name that cannot exist at regular intervals. Nothing legitimate will answer. Any reply gives you the poisoner’s IP and MAC address to alert on.
  • Follow the authentication. A poisoned machine authenticates to an address that is not a server. Look for NTLM network logons (event 4624, logon type 3) and NTLM validations on domain controllers (event 4776) that name an unexpected workstation, and for outbound SMB or HTTP authentication from clients to other clients.
  • Watch the setting itself. Alert when the EnableMulticast policy value changes, since ATT&CK also lists registry changes that turn multicast name resolution back on.
  • Audit yourself. Responder’s analyze mode (-A) shows NBT-NS, mDNS and LLMNR requests without responding. Run by your own team on your own network, it shows which machines still send these lookups after the policy is applied.

How we use it in an engagement: Responder in a penetration test

Responder is an early-stage credential-access tool in internal network penetration testing and red team assessments. It frequently provides the initial credentials in an assumed-breach or on-site test, from which lateral movement begins. It answers the question, “if someone got a connection to your internal network, how quickly could they get a valid credential?”, and the honest answer, on an unhardened network, is very. When the domain itself is the question, an Active Directory security assessment runs this check alongside the other paths to Domain Admin.

On our tests it runs only on the network segments in scope and during the agreed testing windows. Any captured credential is used only to show how far it reaches within the rules of engagement, and the report pairs the finding with the Group Policy and network settings that close it, which the free retest then confirms. Internal tests start at $6,000; see the internal penetration testing price page.

Frequently asked questions

Is Responder legal to run on my own network? Running it on a network you own or are authorized in writing to test is legal security testing. Without that authorization it is capturing other people’s credentials, which computer-crime laws treat as an offense. Inside a company, get written approval first even for analyze mode, because the tool sees other people’s authentication attempts.

Is there a Responder for Windows? Responder is a Python tool, and its README lists Linux as the requirement, with macOS supported. For Windows hosts, testers usually reach for Inveigh, which its author describes as a cross-platform .NET machine-in-the-middle tool for penetration testers. The defenses are the same for both.

Does disabling LLMNR break anything? Rarely, where DNS works. What breaks is anything that only ever resolved by short name on the local subnet, typically an old device or a share reached by a typed nickname. A pilot group finds these before the full rollout.

Does SMB signing stop Responder? It stops the relay half. With signing required, a captured authentication cannot be relayed to SMB on that server. It does not stop the capture, so the hash can still be cracked offline if the password is weak. You need both fixes.

What is Responder analyze mode? A listen-only mode, started with -A, in which Responder shows the NBT-NS, mDNS and LLMNR requests on the segment without answering any of them. Defenders use it to measure how much of this traffic a network produces.

The short version

Responder captures Windows credentials by answering the unauthenticated LLMNR and NBT-NS requests that default Windows networks send whenever a name lookup fails. It needs no credentials and no exploit, waits for failed lookups and answers them, and often yields a domain credential within minutes, the foothold for everything after. The defense is decisive: disable LLMNR and NBT-NS, enforce SMB signing, and use strong passwords. It is one of the most common findings on internal tests precisely because the fix is easy and so often missed.

Curious how quickly someone with a connection to your network could capture a working credential? Finding that out is a standard part of every internal penetration test we run. Scope one here.

Mark Kiss

Written by

Mark Kiss, Founder & CEO

15+ years in the field. Focus: Offensive security & strategy. Part of the team that scopes, tests, and reports for our clients. Full profile →

All articlesRed Teaming

Find out what an attacker sees.

Tell us what to test and see your fixed price.

Prefer the full scoping questionnaire? 
Start the conversation

Get a quick proposal

We are online