

How Pass-the-Hash Attacks Work and How Defenders Detect Them

Pass the hash is an attack in which someone who has stolen a Windows account’s NT hash uses it to sign in as that account over NTLM, without ever learning the password. The technique persists wherever Windows systems still accept NTLM and stolen credentials keep access to other hosts.
Password complexity slows guessing, but it offers little protection once an attacker has stolen the password-derived value that NTLM uses for authentication. The resulting logon can look ordinary because the attacker computes a valid authentication response. Defenders have to establish where the credential came from, which system used it, and whether the account should have reached the destination.
TL;DR:
- Possession of an NT hash is enough to compute NTLM authentication responses, including NTLMv2 responses.
- Detection requires correlation across source-host, target-host, and credential-validation telemetry.
- Password changes address exposed credentials but leave existing access and attacker persistence to investigate.
- Windows credential protection, unique local passwords, and authentication restrictions reduce different parts of the exposure.
Why a Stolen NTLM Hash Works Like a Password
Both NTLMv1 and NTLMv2 response calculations start from the NT hash, an MD4 digest of the password’s UTF-16LE encoding. NTLMv2 adds the username and domain to derive its response key, but the secret underneath is still the NT hash. An attacker holding it can answer any fresh challenge without recovering the password.
During authentication, the server sends a challenge and the client answers with a response built from that password-derived material. The stored NT hash itself never travels across the network.
Captured NetNTLM challenge-response material is a different artifact. It contains proof tied to an authentication exchange, not the reusable NT hash. Attackers may try offline password guessing against it or relay an exchange under suitable conditions.
Password cracking and brute force both work toward recovering the plaintext by testing candidate passwords. Pass the hash skips that recovery step entirely.
How a Pass-the-Hash Attack Unfolds
A typical chain moves from a compromised endpoint toward domain administrator access through repeated credential theft and reuse. Each move is bounded by the account’s permissions, reachable services, and host protections, so domain admin is a possible outcome, not a given.
Gaining Local Admin or SYSTEM Access
Reading LSASS and other live credential stores usually takes local admin or SYSTEM access. An attacker gets it through privilege escalation, compromised administrative credentials, or an already privileged service. Two routes skip this step: delegated directory-replication rights let an attacker run DCSync for domain hashes without local admin on a domain controller, and stolen backups or earlier dumps allow offline extraction with no live privileged session.
Dumping Hashes From LSASS, SAM, and NTDS.dit
LSASS memory can hold NT hashes and Kerberos tickets for active sessions, depending on how the user signed in, but it won’t always hold a plaintext password. Credential Guard and LSA protection limit what an attacker can pull from it.
The local SAM holds local-account password hashes, while NTDS.dit holds domain-account credential data on domain controllers. Cached domain logon verifiers (DCC2) are a separate case: they can be cracked but cannot be passed.
Mimikatz can extract accessible LSASS material and analyze dumps. Impacket secretsdump supports offline hive and NTDS parsing and remote collection methods, including directory replication. Attackers can also collect from memory dumps, registry exports, and backups.
Authenticating With the Stolen Hash
A stolen hash carries no extra privileges. It can authenticate an SMB session with only the account’s existing rights, so PsExec-style execution still needs service-control rights and typically a writable share. WMI execution needs DCOM connectivity, namespace permissions, firewall rules, and administrative rights.
Mimikatz’s sekurlsa::pth starts a process under placeholder credentials, then swaps in the stolen hash. Only that process’s outbound network credentials change, not its local privileges on the source host.
Moving Laterally Toward Domain Admin
Every newly accessed host can expose more credentials. Reused local administrator passwords extend this path because identical passwords produce identical NT hashes.
Investigators conducting broader lateral movement detection should track credential use alongside remote services, execution, and newly accessed systems.
Where Pass the Hash Shows Up in Real Intrusions
In recent public cases, pass the hash is how ransomware operators get from a foothold to privileged systems. MITRE ATT&CK tracks it as T1550.002.
CISA reported in August 2026 that Gunra ransomware actors ran secretsdump against compromised domain controllers to extract NTDS password hashes, then used pass-the-hash or pass-the-ticket methods to move laterally into other privileged systems.
In a BlackSuit intrusion that began in December 2023, The DFIR Report found type 9 logons with the logon process seclogo, followed by a move from SYSTEM into a domain administrator context and renewed Cobalt Strike lateral movement. The sequence, not the event fields alone, pointed to pass the hash.
A June 2024 ELPACO-team intrusion began with an exploited internet-facing Confluence server. The actor harvested credentials with Mimikatz, ProcessHacker, and Impacket secretsdump, tested hashes against two domain controllers, and then ran wmiexec with an NTLM hash argument as a domain admin. Ransomware followed about 62 hours after initial exploitation.
In the two DFIR Report intrusions, the tell was a chain: credential access, then hash-based authentication, then privileged activity. The detection guidance below correlates that whole chain.
Pass the Hash vs Pass the Ticket, Overpass the Hash, and NTLM Relay
The authentication protocol and reused artifact distinguish these techniques more reliably than tool names:
MITRE ATT&CK tracks ticket reuse as a separate technique from hash-based authentication. Reusing a TGT may still require contacting the KDC to obtain service tickets; reusing an existing service ticket may not.
“Pass the key” is practitioner terminology for Kerberos authentication using a stolen long-term key, often AES. Usage varies, so record the key type and ticket-request behavior.
How to Detect Pass-the-Hash Activity
Detection depends on correlating source-host activity, target-host logons, and credential validation. An NTLM logon event shows only that NTLM was used; the surrounding evidence decides whether that use was suspicious.
Windows Event Log Signals
On the target host, Event 4624 commonly records NTLM network authentication with LogonType=3, AuthenticationPackageName=NTLM, and LogonProcessName=NtLmSsp.
KeyLength records the NTLM session-security key length, and a zero value appears in some pass-the-hash detection rules. Zero also appears on Kerberos logons, and legitimate NTLM logons can share the other fields, so treat KeyLength=0 as a filter rather than proof.
Source-host Event 4624 with LogonType=9, LogonProcessName=seclogo, and AuthenticationPackageName=Negotiate is associated with Mimikatz-style credential substitution. A Sigma rule written to detect overpass the hash ties this pattern to Mimikatz’s sekurlsa::pth module, the same command attackers use for pass the hash. Legitimate runas /netonly produces the same pattern, and other pass-the-hash methods may not, so absence does not clear a host.
Event 4776 gives the hunter a record of NTLM credential validation on the authoritative computer: a domain controller for domain accounts, the local computer for local accounts. It names the source workstation but not the destination resource. Local accounts are validated on the local computer, so reusing one leaves no domain-controller 4776.
Event 4672 shows which new logons were assigned sensitive privileges, and it is most useful when you correlate its Logon ID with 4624 on the same host. On its own it doesn’t prove privilege escalation or malicious activity.
Sysmon Event 10 records process access and can expose suspicious access to lsass.exe, where access masks and call traces narrow the investigation, though no single mask uniquely identifies Mimikatz or confirms extraction.
Match collection to the hunt: enable successful Audit Logon for 4624, Audit Credential Validation for 4776, and Audit Special Logon for 4672, and add failure auditing if you also want rejected attempts. Sysmon needs a ProcessAccess configuration that includes relevant LSASS access.
To cover the share access in the worked example below, turn on Audit File Share for 5140 and Detailed File Share for 5145, since neither is on by default. Detailed File Share logs every file access; File Share logs one event per connection. If events are still missing, the cause may be disabled auditing, filtering, absent forwarding, or a different attack implementation.
A Worked Example in the Logs
Picture a hash dumped from LSASS on host A and reused over SMB against host B. The hunt tests credential access and the later authentication separately, and the table shows the records each step typically leaves. Exact records vary with your audit policy and tooling. Atomic Red Team’s open-source tests for this technique let you run the attack in a lab and capture the paired source, target, and domain-controller records from your own environment.
Host B’s 5145 Logon ID links share-access checks to its 4624, but Logon IDs are host-local and should not be joined across computers. Cross-host correlation instead relies on account identifiers, timestamps, and available network fields, which may be incomplete or client-supplied.
Escalate the sequence when suspicious LSASS access precedes unexpected account use and remote execution. A successful SMB logon alone leaves the credential’s origin unresolved.
Tuning Out False Positives
Baseline each legitimate NTLM source before you suppress it. If a legacy application routinely uses NTLM, hunt changes in its account, source, destination, and schedule.
If a service account normally authenticates from designated servers, investigate use elsewhere. Machine accounts also deserve scrutiny; a trailing dollar sign does not establish benign behavior.
If administrators use runas /netonly, correlate type 9 logons with the initiating process, approved work, and destinations. If security software accesses LSASS, validate its signer, binary, deployment, and behavior before creating an exception.
Keep deviations from approved patterns as findings even when the account is a trusted administrator, and feed them into identity threat investigation.
How to Prevent and Mitigate Pass the Hash
Each control below closes a different part of the exposure, so apply them together and test compatibility at each enforcement point.
- Credential Guard protects selected domain secrets through virtualization-based security. It is on by default for eligible Windows 11 22H2 and later devices, and for eligible domain-joined Windows Server 2025 systems that aren’t domain controllers, though licensing, hardware, and prior explicit disablement affect deployment. It does not protect SAM or NTDS.dit.
- LSA protection, or RunAsPPL, restricts untrusted LSASS memory access and code loading. New Windows 11 22H2 and later installs on enterprise-joined, HVCI-capable devices turn it on automatically. On upgraded Windows 11 devices, don’t assume it turned itself on; check that LSASS is running as a protected process, for example through WinInit event 12. On Windows Server 2025, turn it on and verify it explicitly.
- Windows LAPS creates and rotates unique managed local administrator passwords to limit cross-host reuse. Password retrieval permissions and backup protection remain security boundaries.
- Administrative separation keeps control-plane credentials away from lower-trust systems. Dedicated accounts and restricted sign-in paths reduce opportunities to harvest more privileged credentials, and privileged access management with just-in-time elevation leaves fewer standing admin credentials in memory to steal.
- Protected Users blocks NTLM and weaker Kerberos preauthentication for eligible domain users under the required domain configuration. Test AES availability, delegation dependencies, and offline-sign-in requirements, and keep service and computer accounts out of the group.
- NTLM restrictions block hash-based NTLM authentication where enforced. Audit dependencies before denying traffic. Newer Windows releases can also block NTLM on the SMB client alone, which covers a narrower path than organization-wide restrictions.
- Denying network logon to local accounts blocks remote use by account type. Choose between all local accounts (S-1-5-113) and administrative local accounts (S-1-5-114) carefully: denying every local account can break failover clustering, which relies on the nonadministrative CLIUSR account, while S-1-5-114 keeps cluster services working.
Credential protection makes hashes harder to steal from a protected host, but a hash stolen somewhere else still authenticates wherever NTLM is accepted. The authentication restrictions matter as much as the theft protections.
NTLM still works despite deprecation. Microsoft has deprecated all NTLM versions, and NTLMv1 is removed starting with Windows 11 24H2 and Windows Server 2025, but the removal is partial: MS-CHAPv2 still relies on the NTLMv1 response function, and Credential Guard stops MS-CHAPv2 from using signed-in credentials for single sign-on. Since the October 13, 2026 update, Windows also blocks NTLMv1-derived SSO credentials by default on devices where Credential Guard is off and BlockNtlmv1SSO is not set.
Under Microsoft’s January 2026 roadmap, enhanced NTLM auditing is available now, and IAKerb and Local KDC, still pre-release, are planned for Windows Server 2025 and Windows 11 24H2 and later in the second half of 2026. Network NTLM is slated to be disabled by default starting with the next major Windows Server release and its matching client releases, though it can be re-enabled by policy and Microsoft says the timeline may change. No shipping version disables network NTLM by default yet.
Stolen Hashes Are an Investigation Problem as Much as a Hardening One
Hardening narrows the exposure, but a reused hash still produces a logon someone has to explain. Investigators need identity permissions, endpoint activity, prior authentication patterns, and approved administrative work to distinguish credential reuse from expected access. Daylight’s Threat Hunting engagements, run separately from MDR, are built to test that kind of credential-reuse hypothesis against identity, endpoint, and historic telemetry.
Scope the incident before acting: establish the exposed accounts and hosts, preserve evidence, and assign containment actions as part of incident response triage. Resetting one account before establishing the scope can leave other stolen credentials and persistence untouched.
Frequently Asked Questions About Pass the Hash
Does Changing the Password Stop Pass the Hash?
Only partly, because changing an account’s password replaces its NT hash but does not remove the attacker’s foothold. For domain users, the old password keeps working for NTLM network access for five minutes by default, and existing sessions or tickets need separate handling. If NTDS.dit or krbtgt secrets were exposed, treat it as domain compromise: Microsoft’s forest recovery procedure is to reset krbtgt twice, waiting at least the maximum ticket lifetime (10 hours by default) between resets. An isolated hash theft doesn’t call for a krbtgt reset.
Does MFA Prevent Pass the Hash?
MFA alone does not prevent pass the hash. NTLM does not support MFA and can be misused to bypass MFA requirements. Joint CISA, NSA, and ASD guidance therefore recommends disabling NTLM wherever possible. Most MFA at an interactive sign-in, VPN, or identity-provider entry point adds no second-factor challenge to later NTLM network authentication, because NTLM validates a password-derived response. Blocking NTLM on the path to the resource is what closes that gap.
Does Kerberos-Only Authentication Stop Pass the Hash?
It stops classic pass the hash on any path where NTLM is fully blocked, including fallback and exceptions, but it doesn’t end hash reuse. Where a domain still accepts RC4 for Kerberos, the same stolen NT hash works as a Kerberos key for overpass the hash. Even with RC4 disabled, stolen AES keys still support pass the key.
Can Pass the Hash Work Against Microsoft Entra ID?
The hash Entra ID stores gives an attacker nothing to pass. Password hash sync sends Entra ID a salted PBKDF2 hash derived from the NT hash, and that stored value can’t be passed back to on-premises Active Directory. The exposure returns with Microsoft Entra Domain Services, which can sync NTLM password hashes into its managed domain for legacy applications. Turning that sync off removes them, though legacy applications that depend on NTLM can break.
Can Pass the Hash Work Outside Windows?
Yes. Linux clients can use Impacket or Samba’s smbclient to authenticate to Windows services with an NT hash in place of a password. Acceptance depends on the target’s NTLM configuration and account permissions. A Linux source has no Windows seclogo or lsass.exe footprint, so hunt target-side authentication and validation records. Samba servers also participate in NTLM authentication, subject to their local and domain policies.
Subscribe to newsletter
Subscribe to receive the latest blog posts to your inbox every week.
.avif)




Related posts
.png)

For expert judgment, scaled to your world, head for Daylight



