← Back to the cybersecurity blog

Protecting LSASS from credential dumping on Windows: what to enable, how to verify it and what to monitor

By Daniel O'Grady · CIO y socio fundador · Published: 15 November 2026 · Updated: 15 November 2026
Protecting LSASS from credential dumping on Windows

When a domain administrator signs in to a laptop to deal with a support call, Windows keeps enough in the memory of lsass.exe for an attacker to reuse that identity on other machines: the NTLM hash of the password and a set of Kerberos tickets. The process is the Local Security Authority Subsystem Service (LSASS). Dumping its memory is a routine step in intrusions that spread laterally and end in ransomware.

In 2022 Microsoft listed the methods it had seen between March and August of that year: the MiniDump export of comsvcs.dll launched through rundll32, Mimikatz and its forks, procdump, and even Task Manager. All of them begin with a privileged process opening lsass.exe with the right to read its memory. For years Windows has shipped three separate barriers against this, plus one legacy setting that needs watching. It is common to find estates where one of the barriers is missing and the others have never been verified.

The guide is aimed at infrastructure leads and SOC analysts in mid-sized organisations whose Windows devices are joined to Active Directory or Microsoft Entra ID and managed through Group Policy or Intune. For each barrier it gives the exact value that enables it and a way to prove it is enforced, and, for the barriers controlled through the registry, a query that catches someone switching them off.

What does LSASS hold, and why dump it?

It holds the credentials of whoever has signed in to the device. The Local Security Authority validates sign-ins and enforces the device's security policy. So that users are not asked for their password every time they open a file share or an internal application, the authentication packages it loads (NTLM, Kerberos and, on old or misconfigured systems, WDigest) keep the secrets of every active session in memory. Attackers exploit precisely that convenience: almost any account that signs in interactively leaves its secrets in that process while the session is open, and sometimes after it ends.

Dumping LSASS sits in the middle of an intrusion. The attacker first needs local administrator rights, usually after privilege escalation. Next comes a request to Windows for a handle to the process with memory-read rights, a copy of that memory to disk or a live read, and extraction of the secrets, either on the spot or, from a dump file, offline on another machine. MITRE ATT&CK catalogues this as T1003.001, LSASS Memory, and its detection strategy (DET0363) follows that sequence: broad access to lsass.exe, then a dump file or a registry change.

The attacker leaves traces in three places: process-access telemetry (Sysmon event 10, or its equivalent from an EDR), new .dmp files, and registry edits that weaken protection. The detection section has a query for each.

Which barriers does Windows offer against dumping?

What does LSA protection (RunAsPPL) do?

With this option LSASS starts as a Protected Process Light (PPL). It loads only code signed to Microsoft's requirements, and an ordinary process, even one running as administrator, cannot open it to read memory or inject code. The protection defeats user-mode dumping tools which open LSASS with a standard handle.

Kernel-mode code is another matter. An attacker who manages to run code at that level, for instance by loading a signed but vulnerable driver, operates above the protection and can remove it. It should be paired with the vulnerable driver blocklist, memory integrity (HVCI) and, where Defender is in use, the ASR rule that blocks abuse of exploited vulnerable signed drivers.

How far does Credential Guard go?

Credential Guard uses virtualisation-based security to move secrets into an isolated process, LsaIso.exe, which not even the Windows kernel can read directly. Microsoft says it protects NTLM password hashes, Kerberos ticket-granting tickets (TGTs) and credentials that applications store as domain credentials.

Microsoft also lists what it does not cover: the Active Directory database and the SAM database that holds local accounts. It is not recommended on domain controllers and not supported on Exchange Server. From Windows 11 22H2 and Windows Server 2025 it is on by default for devices with Enterprise E3/E5 or Education A3/A5 licences (or domain-joined Windows Server that is not a domain controller), provided the hardware qualifies and nobody has explicitly disabled it. Windows Pro does not support it.

Where does the ASR rule against credential theft fit?

The attack surface reduction rule "Block credential stealing from the Windows local security authority subsystem" blocks memory access to LSASS from unauthorised processes. It needs Microsoft Defender Antivirus running in active mode, belongs to the standard protection set and has no warn mode. Microsoft warns that audit mode generates a high volume of events, most of them safe to ignore, and that once LSA protection is enabled the rule adds no further protection and the portal reports it as not applicable.

Why does WDigest still matter?

Because an attacker can switch it back on. WDigest is a legacy protocol that used to hold the user's password unencrypted in memory. From Windows 8.1 and Server 2012 R2 the system behaves as if UseLogonCredential were 0, and the password is not stored; on Windows 7, 8, 2008 R2 and 2012 with update KB2871997 the value has to be created and set to 0.

Microsoft has documented attackers who flip the value to 1, for example when a dump is not enough, so the next sign-in leaves a cleartext password behind. Few legitimate processes ever make that change, which makes it a high-fidelity signal.

Which combination suits each device type?

The three barriers work together. LSA protection is the foundation because it runs on almost any modern Windows build; Credential Guard adds isolation where licensing and hardware allow; the ASR rule covers devices where process protection cannot yet be enabled.

How the LSASS protections compare
ControlWhat it protectsPrerequisitesCompatibility riskHow to roll back
LSA protection (RunAsPPL)Stops unprotected processes from opening LSASS to read memoryWindows 8.1 or Server 2012 R2 onwards; UEFI and Secure Boot for the UEFI lockLSA plug-ins or drivers without a Microsoft signature (smart card, password filters) stop loadingPolicy or registry; with UEFI lock, the LsaPplConfig.efi tool on each device
Credential GuardIsolates NTLM hashes, Kerberos TGTs and stored domain credentialsVBS and Secure Boot; TPM recommended; Enterprise, Education or Windows ServerRemoves single sign-on for NTLMv1, MS-CHAPv2 (Wi-Fi or VPN) and CredSSP; blocks Kerberos DES and unconstrained delegation; saved Windows credentials in Remote Desktop stop workingPolicy; with UEFI lock, physical presence at boot
ASR rule 9e6c4e1f-7d60-472f-ba1a-a39ef669e4b2Blocks memory access to LSASSMicrosoft Defender Antivirus in active modeLimited exclusions; known issue with Quest Dirsync Password SyncChange the rule mode in Intune or Group Policy
WDigest with UseLogonCredential set to 0Keeps cleartext passwords out of memoryDefault behaviour since Windows 8.1Single sign-on to apps using Digest authenticationRegistry or policy

Sources: Microsoft Learn (LSA protection, updated 26 March 2025; Credential Guard, 27 April 2026; Credential Guard known issues, 23 April 2026; ASR rules reference, 9 September 2026). Checked on 15 November 2026.

By device type, that leaves five main cases. The same logic applies to Windows 10 Enterprise and to member servers running 2016 to 2022, except that only RunAsPPL value 1 enables protection there. For endpoints and member servers we recommend starting without the UEFI lock for the first cycle.

Recommended combination by device type
Device typeCombinationNote
Windows 11 Enterprise or Education endpointLSA protection, Credential Guard, WDigest at 0The ASR rule is reported as not applicable once LSA protection is on
Windows Pro endpointLSA protection and WDigest at 0; ASR rule in block mode until LSA protection is confirmed activeCredential Guard is unavailable on this edition
Windows Server 2025 member serverLSA protection and Credential GuardOn Hyper-V hosts, move live migration to Kerberos constrained delegation first
Domain controllerLSA protection with UEFI lock after auditing every plug-inReview password filters (the Notification Packages value under the Lsa key); Credential Guard adds no security here
Exchange serverLSA protection after auditing plug-insCredential Guard is unsupported

Hard2bit analysis based on Microsoft Learn documentation.

UEFI lock on domain controllers is consistent with the CIS Microsoft Windows Server 2025 Benchmark v2.0.0, item 18.9.27.2 (accessed via Tenable), which expects "Enabled with UEFI Lock" in its NG profile. CIS makes the identical recommendation for member servers, so leaving them unlocked at first is a deliberate departure: if an incompatible plug-in surfaces after the lock is in place, rolling back means visiting each device.

A mistake on a domain controller costs more still, because reversal needs physical or hypervisor console access and Microsoft's opt-out tool. Prior auditing there is mandatory.

How is each barrier rolled out?

Where should the rollout start?

With an audit of LSA protection compatibility, to find out which plug-ins would fail before enforcement. Microsoft documents an audit mode: under HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution Options\LSASS.exe, create a REG_DWORD named AuditLevel with data 8 and restart.

Events 3065 and 3066 in Microsoft-Windows-CodeIntegrity/Operational name each module that would have been blocked. With Smart App Control on, these events are not written. On Windows 11 22H2 and later this audit mode is on by default, so the events may already exist without creating the value. After enforcement, events 3033 and 3063 in that log identify modules that were refused.

How is LSA protection enabled?

The Group Policy setting is under Computer Configuration, Administrative Templates, System, Local Security Authority, named "Configures LSASS to run as a protected process", with the options "Enabled with UEFI Lock" and "Enabled without UEFI Lock".

The registry equivalent is RunAsPPL under HKLM\SYSTEM\CurrentControlSet\Control\Lsa: 1 enables it with UEFI lock where UEFI and Secure Boot are present (registry-only otherwise), 2 enables it without the lock, and 0 disables it. Only Windows 11 22H2 and later honour the value 2; on Windows 10 and Windows Server 2016 to 2022, only 1 provides protection.

Microsoft's guidance flags two traps. Returning the policy to Not Configured does not remove protection already applied; the disable option has to be chosen explicitly. And with UEFI lock the setting is stored in a firmware variable, so removing it requires the LSA Protected Process Opt-out tool (LsaPplConfig.efi).

On Windows 11 22H2 and 23H2, protection switches on by itself for clean installs joined to a domain or to Entra ID that support HVCI, always without the lock. From Windows 11 24H2, Microsoft says upgraded devices also get it, after a five-day evaluation period and a reboot, provided no incompatibilities are detected; a configured enterprise policy overrides that behaviour. The plug-in audit therefore matters even where nobody has deployed the setting. Automatic enablement does not always write the value to the registry either, and the boot event described below is the most dependable proof.

Where is Credential Guard enabled?

Intune's Settings catalog exposes it under the Device Guard category as Credential Guard, with "Enabled with UEFI lock" or "Enabled without lock". In a custom profile, the DeviceGuard configuration service provider takes EnableVirtualizationBasedSecurity set to 1 and LsaCfgFlags set to 1 (locked) or 2 (unlocked). If someone explicitly disabled it before the move to Windows 11 22H2, default enablement leaves it off, so an inventory of where it is running is needed.

Three known incompatibilities need checking before a wide rollout, all documented by Microsoft. Wi-Fi or VPN using PEAP-MSCHAPv2 loses single sign-on, and Microsoft recommends moving to certificate-based EAP-TLS or PEAP-TLS. Windows credentials saved in the Remote Desktop client stop working. On Windows Server 2025, Hyper-V live migration over CredSSP delegation fails when the source host runs Credential Guard.

In which mode should the ASR rule start?

In audit mode on a small pilot group, moving to block once the processes that appear have been reviewed; Microsoft also allows going straight to block on a small group, because audit mode for this rule is noisy. The rule's GUID is 9e6c4e1f-7d60-472f-ba1a-a39ef669e4b2. Seeing svchost.exe in the events is normal; blocking its memory reads does not stop the process from working.

Can enforcement be proved device by device?

Yes, with checks that only read device state, since enabling without checking is how organisations end up believing they are protected when they are not:

# RunAsPPL records intent: 1 locked where Secure Boot is present, 2 unlocked (Windows 11 22H2 or later only), 0 disabled.
Get-ItemProperty -Path 'HKLM:\SYSTEM\CurrentControlSet\Control\Lsa' -Name RunAsPPL -ErrorAction SilentlyContinue | Select-Object RunAsPPL
# The dependable proof is the Wininit events since the last boot: 12 for protected LSASS, 13 to 17 for Credential Guard.
$boot = (Get-CimInstance Win32_OperatingSystem).LastBootUpTime
Get-WinEvent -FilterHashtable @{LogName='System'; ProviderName='Microsoft-Windows-Wininit'; Id=12,13,14,15,16,17; StartTime=$boot} -ErrorAction SilentlyContinue | Select-Object TimeCreated, Id, LevelDisplayName, Message

# Credential Guard is running if the list contains the value 1.
(Get-CimInstance -ClassName Win32_DeviceGuard -Namespace root\Microsoft\Windows\DeviceGuard).SecurityServicesRunning -contains 1
# ASR rule mode: 0 disabled, 1 block, 2 audit, 5 not configured; this rule does not support 6 (warn).
$g = '9e6c4e1f-7d60-472f-ba1a-a39ef669e4b2'; $p = Get-MpPreference; $ids = @($p.AttackSurfaceReductionRules_Ids); $a = @($p.AttackSurfaceReductionRules_Actions)
$i = -1; for ($n = 0; $n -lt $ids.Count; $n++) { if ($ids[$n] -eq $g) { $i = $n } }; if ($i -ge 0) { $a[$i] } else { 'Rule not configured' }

# The rule only acts when Defender runs in active mode: AMRunningMode should read Normal, not Passive.
Get-MpComputerStatus | Select-Object AMRunningMode, RealTimeProtectionEnabled, IsTamperProtected
# WDigest: should be 0 or absent on modern systems.
Get-ItemProperty -Path 'HKLM:\SYSTEM\CurrentControlSet\Control\SecurityProviders\WDigest' -Name UseLogonCredential -ErrorAction SilentlyContinue

Event 12 should state that LSASS started as a protected process with level 4. Event 13 confirms that Credential Guard started and event 14 records its configuration. Event 15 means it is configured but the secure kernel is not running, for example because virtualisation is disabled in firmware; event 16 means LsaIso failed to launch, and event 17 points to an error reading the UEFI configuration.

Microsoft advises against checking for LsaIso.exe in Task Manager; msinfo32 shows the answer on the line for virtualisation-based security services running. Once Credential Guard is active, event 4013 in Microsoft\Windows\NTLM\Operational records blocked NTLMv1 attempts.

Which queries detect dumping attempts?

Four Microsoft Defender XDR queries and a Sigma rule for Sysmon. Hardening narrows the attack surface, yet the SOC still has to see blocked attempts, attempts to disable the barriers and dumps that arrive by an uncovered route. The queries run in Microsoft Defender XDR advanced hunting and need devices onboarded to Defender for Endpoint Plan 2 (included in Microsoft 365 E5). The Sigma rule covers teams sending Sysmon data to another SIEM.

Has the ASR rule blocked or audited any attempts?

// LSASS access blocked or audited by the ASR rule over the past 7 days.
// Action types come from the official ASR rules reference.
DeviceEvents
| where Timestamp > ago(7d)
| where ActionType in ("AsrLsassCredentialTheftBlocked", "AsrLsassCredentialTheftAudited")
// Grouping by source process separates recurring noise from new activity.
| summarize Events = count(), Devices = dcount(DeviceName), First = min(Timestamp), Last = max(Timestamp) by InitiatingProcessFileName, InitiatingProcessFolderPath, ActionType
| order by Devices asc, Events asc

Sorting by device count, lowest first, brings to the top whatever appears on one or two machines. Signed system processes seen across the estate (svchost.exe in System32, WmiPrvSE.exe in System32\wbem, MsMpEng.exe under ProgramData\Microsoft\Windows Defender\Platform) are known noise, as long as the full path and signer check out. An executable running from a temp or user folder needs investigating straight away.

Is anyone trying to switch the barriers off?

// Registry changes that weaken LSASS protection, including deleted values.
// has and endswith are used because the path may appear as HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001.
DeviceRegistryEvents
| where Timestamp > ago(30d)
| extend Data = tostring(RegistryValueData), Deleted = ActionType endswith "Deleted"
// WDigest.
| where (RegistryKey has @"\Control\SecurityProviders\WDigest" and RegistryValueName =~ "UseLogonCredential" and Data in ("1", "0x00000001"))
// LSA.
or (RegistryKey endswith @"\Control\Lsa" and RegistryValueName in~ ("RunAsPPL", "LsaCfgFlags") and (Data in ("0", "0x00000000") or Deleted))
// VBS.
or (RegistryKey has @"\DeviceGuard" and RegistryValueName in~ ("LsaCfgFlags", "EnableVirtualizationBasedSecurity") and (Data in ("0", "0x00000000") or Deleted))
| project Timestamp, DeviceName, ActionType, RegistryKey, RegistryValueName, PreviousRegistryValueData, RegistryValueData, InitiatingProcessFileName, InitiatingProcessAccountName, InitiatingProcessCommandLine

Very few hits are expected. Legitimate changes should come from the management agent (Intune or the Group Policy client) during known change windows; an edit made with reg.exe or PowerShell under a user account deserves a high-priority alert. Microsoft does not document how RegistryValueData renders DWORD values, hence the two accepted formats. SigmaHQ maintains comparable rules for WDigest and for disabling Credential Guard.

Are built-in tools being used to dump LSASS?

// comsvcs.dll or procdump aimed at LSASS, including renamed binaries.
DeviceProcessEvents
| where Timestamp > ago(30d)
| where ((FileName =~ "rundll32.exe" or ProcessVersionInfoOriginalFileName =~ "rundll32.exe") and ProcessCommandLine has "comsvcs" and ProcessCommandLine contains "minidump")
// procdump variant, renamed binaries included.
or (ProcessCommandLine has "lsass" and (FileName in~ ("procdump.exe", "procdump64.exe", "procdump64a.exe") or ProcessVersionInfoCompanyName has "Sysinternals"))
// Columns showing who launched the process.
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine

The process hunt covers the stage where an attacker leans on legitimate binaries, a pattern explored in our guide to detecting abuse of trusted Windows binaries. It has known blind spots: procdump launched against a process ID instead of the name, and comsvcs invoked by ordinal; the latter is covered by SigmaHQ's Process Memory Dump Via Comsvcs.DLL rule. Support tools that capture diagnostic dumps can trigger it; they are recorded as exceptions by path and signer, never by file name.

Are there dump files on disk?

// .dmp files written by processes linked to dumping, or with lsass in the name.
DeviceFileEvents
| where Timestamp > ago(30d)
| where ActionType == "FileCreated" and FileName endswith ".dmp"
// lsass in the name or a process linked to dumping.
| where FileName has "lsass" or InitiatingProcessFileName in~ ("rundll32.exe", "taskmgr.exe", "procdump.exe", "procdump64.exe", "procdump64a.exe")
// File path and the process that created it.
| project Timestamp, DeviceName, FolderPath, FileName, InitiatingProcessFileName, InitiatingProcessCommandLine, InitiatingProcessAccountName

Task Manager writes legitimate dumps when an engineer requests one for troubleshooting, and Windows Error Reporting creates its own after a crash. With comsvcs the attacker picks the file name and extension, so dumps saved under another extension are missed here; the process hunt covers that gap. A dump of lsass.exe with no open support case behind it should be handled as an incident.

Does Sysmon work as an alternative to Defender?

Yes. Without Defender, Sysmon event 10 (process access) is the equivalent source, provided the Sysmon configuration records access to lsass.exe and captures the CallTrace field. The rule below combines two conditions: memory-read access from anything other than known agents, and access from rundll32.exe or taskmgr.exe whose call stack runs through the dump libraries. It is written in compact YAML, which does not depend on indentation:

title: Memory-read access to LSASS outside known agents
id: 7c2e9d41-0b8a-4f63-a5d2-91e4c6f08b37
status: experimental
description: Memory-read access to lsass.exe from outside the approved agents, or a dump from rundll32 or taskmgr.
references: ['https://attack.mitre.org/detectionstrategies/DET0363']
author: Hard2bit
date: 2026-11-15
logsource: {category: process_access, product: windows}
# Agent paths: adjust to the estate.
detection: {selection: {TargetImage|endswith: '\lsass.exe', GrantedAccess: ['0x1010', '0x1410', '0x1F0FFF', '0x1FFFFF']}, filter_agents: {SourceImage: ['C:\Windows\System32\svchost.exe', 'C:\Windows\System32\wbem\WmiPrvSE.exe']}, selection_lolbin: {TargetImage|endswith: '\lsass.exe', SourceImage|endswith: ['\rundll32.exe', '\taskmgr.exe'], CallTrace|contains: ['comsvcs.dll', 'dbghelp.dll', 'dbgcore.dll']}, condition: '(selection and not filter_agents) or selection_lolbin'}
# Usual false positives.
falsepositives: ['Security or inventory agents without an exception', 'Diagnostic dumps taken with Task Manager']
level: high
tags: [attack.credential-access, attack.t1003.001]

The access masks are a sample. 0x1F0FFF, the value cited in MITRE's DET0363, is the PROCESS_ALL_ACCESS constant inherited from Windows XP and Server 2003; 0x1FFFFF is the same right from Windows Vista onwards; 0x1010 combines PROCESS_QUERY_LIMITED_INFORMATION and PROCESS_VM_READ, and 0x1410 adds PROCESS_QUERY_INFORMATION. Attackers can request other combinations, so this rule complements the queries above without covering every variant.

Exceptions in filter_agents are defined by full path, not by folder, because an attacker with administrator rights can write to Program Files. Elastic tackles this with different logic in its LSASS Process Access via Windows API rule, grouping by normalised path and alerting only when an executable appears on a single host with three accesses or fewer, a low-prevalence approach designed to keep noise down.

Can detection be validated without dumping LSASS?

Yes. Testing the logging pipeline does not involve dumping LSASS. The test needs an isolated lab machine holding no live credentials, with nobody signing in interactively while it runs. Once UseLogonCredential is set to 1 and back to 0, the change should show up in the registry query (or in a custom detection built from it) and, in a SIEM fed by Sysmon, in the SigmaHQ WDigest rule. If no event arrives in the expected window, ingestion is checked first and, if the events are there, the query.

On devices that keep the ASR rule in audit mode, its events from legitimate processes double as a heartbeat for ingestion: if a group of devices goes quiet, something has failed.

Process-access detections are best validated in authorised purple team exercises, in a lab and under change control. It is the principle behind sound detection engineering: a rule that has never been tested against production data only promises coverage.

What are the most common mistakes?

  • The assumption that Credential Guard protects local accounts. The SAM database is out of scope, so local administrator passwords must be unique per device and rotated with Windows LAPS.
  • Credential Guard on domain controllers for consistency's sake. Microsoft advises against it: it adds no security and can cause compatibility problems.
  • The UEFI lock applied across the estate before a plug-in audit. A smart card driver without a valid signature can lock users out, and with the lock in place rollback means touching every device.
  • The ASR rule as the only barrier on devices without LSA protection. It depends on Defender staying in active mode, which an administrator can change unless tamper protection is on.
  • A misread scope: these settings mainly guard the secrets held by LSASS. Browser-stored credentials, DPAPI secrets, the NTDS.dit file on a domain controller and cloud session tokens are all outside their reach, and Microsoft warns that persistent attackers will change technique.
  • Credential Guard rolled out without a review of legacy authentication. It removes single sign-on for NTLMv1, MS-CHAPv2 and CredSSP, so it helps to have done the groundwork on switching off NTLM without breaking the network.

No registry setting fixes the underlying problem. If a domain admin account signs in to a compromised laptop, whoever controls that machine can act through the live session for as long as it lasts, even without extracting the hash.

The answer lies in architecture and the Active Directory tiering model: Tier 0 credentials should never be used on user devices. A decoy account whose credential is deliberately left in memory, as set out in our guide to honeytokens in Active Directory, also means that if an attacker does exploit a dump, the alert fires as soon as that credential is used.

Do these settings count as audit evidence?

Yes. For organisations subject to Spain's National Security Framework (ENS), ISO 27001 or NIS2, these settings produce verifiable evidence. Article 21(2) of NIS2 lists access control policies and basic cyber hygiene among the minimum measures.

Control and evidence mapping
Technical controlENS (RD 311/2022)ISO/IEC 27001:2022, Annex AEvidence
LSA protection and Credential Guard through policyop.exp.2 Security configuration; op.exp.3 Security configuration management8.9 Configuration management; 8.5 Secure authenticationPolicy or Intune profile report plus per-device check output
ASR rule in block modeop.exp.6 Protection against malicious code8.7 Protection against malwareASR rules report from the Defender portal
Active detection queriesop.mon.1 Intrusion detection; op.mon.3 Monitoring8.16 Monitoring activitiesRules deployed in the SIEM and the record of their last validation
Separation of privileged accountsAccess control measures (op.acc)8.2 Privileged access rightsTier 0 inventory and sign-in restrictions

The mapping is indicative; each auditor confirms scope according to system category and the statement of applicability.

If only one thing can be done, what should it be?

An inventory of where LSA protection and Credential Guard are running, based on boot events, not registry values. Plenty of organisations already have part of this protection without knowing it (clean Windows 11 installs, Windows Server 2025 hosts), while others assume they have it even though some device was left out. It might be a 22H2 or 23H2 machine upgraded from Windows 10 or one where a policy prevents it. Everything else follows from that inventory.

The query watching RunAsPPL, LsaCfgFlags and UseLogonCredential should go live the same day: once a barrier blocks a dump, one of the attacker's likely next moves is to try disabling it, and with management agents filtered out, that change stands out clearly.

If in-house staff cannot sustain these detections, a managed SOC service or a focused threat hunting exercise can review existing telemetry for access to LSASS memory. If it turns up, the next step is the incident response plan: isolating the device, resetting the exposed credentials, Tier 0 first, and only then cleaning it up.

Disclaimer: the settings, commands and queries in this guide are offered as technical reference and were checked against Microsoft Learn documentation and the public rules cited as of 15 November 2026. They should be tested in a pre-production environment before being applied, because LSA protection and Credential Guard can prevent plug-ins from loading or break legacy protocols, and the UEFI lock makes rollback harder. The queries need exclusions tuned to each environment. References to Microsoft, Elastic, SigmaHQ, CIS, Quest and other vendors are descriptive and do not constitute a commercial endorsement.

Frequently asked questions

Does Credential Guard replace LSA protection? ▾

No. They work at different layers: LSA protection stops user-mode processes from opening LSASS to read its memory, while Credential Guard moves the secrets into a virtualisation-isolated environment. On Enterprise devices it makes sense to run both, because if Credential Guard fails to start (Wininit events 15 or 16) LSA protection still covers the process.

Can LSASS be protected on Windows Pro devices? ▾

Yes, though without Credential Guard, which requires an Enterprise or Education licence. Windows Pro takes LSA protection (RunAsPPL value 2 on Windows 11 22H2 or later, value 1 on Windows 10), with the ASR rule kept in block mode until event 12 confirms that protection is running.

How can automatic LSA protection on Windows 11 be confirmed? ▾

Checking the registry is not enough, because automatic enablement may not write RunAsPPL. The dependable test is Wininit event 12 in the System log since the last boot, which states that LSASS started as a protected process with level 4. On 22H2 and 23H2 automatic enablement only happens on clean, enterprise-joined installs that support HVCI; from 24H2 it also follows an upgrade if no incompatibilities are detected.

What happens if the UEFI lock is applied and a plug-in stops loading? ▾

Changing the policy will not undo it, because the setting is stored in a firmware variable. Microsoft's LsaPplConfig.efi tool has to be run on every affected device, with console access. That is why auditing first with AuditLevel=8 and reviewing CodeIntegrity events 3065 and 3066 pays off.

Do these measures stop Mimikatz? ▾

They stop variants that read LSASS memory from user mode. They do not stop an attacker who loads a vulnerable driver into the kernel, nor do they protect secrets held outside LSASS, such as the SAM database, NTDS.dit or browser-stored passwords. Hence the need to pair them with registry-change detection and the Active Directory tiering model.

If the SOC can deploy only one query, which should it be? ▾

The one watching registry changes to UseLogonCredential, RunAsPPL, LsaCfgFlags and EnableVirtualizationBasedSecurity. It returns very few results, nearly all of them actionable, and it catches attempts to disable the barriers, a common follow-up when a dump is blocked.

What evidence might an ENS or ISO 27001 auditor ask for on LSASS protection? ▾

The Group Policy or Intune profile report, per-device check output, the ASR rules report from the Defender portal and the list of detection rules with the date each was last validated. Under ENS this maps to op.exp.2, op.exp.3, op.exp.6, op.mon.1 and op.mon.3; under ISO 27001:2022, to Annex A controls 8.5, 8.7, 8.9 and 8.16.

Want a straight answer on scope, priorities and price?

Most organizations reach us mid-question: something needs fixing, nobody has scoped it, and finance wants a number. Thirty minutes with a technical consultant — not a salesperson — gets you a defined scope, priorities ranked by risk and a price range. With what comes out of that call, we turn it into a fixed proposal. We work with organizations in Spain, across the EU and in LATAM.

If your situation is different, tell us anyway — we also take one-off questions on cybersecurity and regulatory compliance.

Based in Spain · Working across the EU and LATAM · ENS High · ISO 27001 · We usually reply in under 24 business hours