← Back to Feed CACHED · 2026-09-10 01:30:15 · CACHE_KEY CVE-2026-83939
CVE-2026-83939 · CWE-822 · Disclosed 2026-09-08

Untrusted pointer dereference in Windows Secure Kernel Mode

ASSESSED — NOISGATE V0.5
Vendor
Reassessed
Verdict:
Do you agree?
01 · The Real Story

Like a bank vault whose walls dissolve — but only after the thief already owns the building

CVE-2026-83939 is an untrusted pointer dereference in Windows Secure Kernel Mode (securekernel.exe), the VTL1 component that enforces Virtualization-Based Security (VBS), Credential Guard, and HVCI. An attacker who already holds SYSTEM / admin-level privileges on the host can manipulate a pointer that the Secure Kernel trusts without validation, escaping the VTL0→VTL1 boundary. The scope-change flag (S:C) in the CVSS vector reflects this: impact crosses a privilege boundary the normal kernel cannot normally reach. Affected builds span Windows 10.0.28000.0 through 10.0.28000.2953, fixed in 10.0.28000.2954. The flaw was disclosed 2026-09-08 as part of a trio of Secure Kernel EoPs (alongside CVE-2026-69846 and CVE-2026-69906) found by researcher vmpr00be via coordinated MSRC disclosure.

Microsoft rated this HIGH at 8.2, and the rating is broadly fair. The S:C scope change is what prevents a downgrade: this isn't just another kernel LPE to SYSTEM — it's a *post*-SYSTEM escalation that defeats the one isolation layer (VBS/Credential Guard) designed to survive kernel compromise. On a domain controller running Credential Guard, success means extracting cached Kerberos TGTs and NTLM hashes that are supposed to be hypervisor-protected. That said, PR:H (high privileges required) and AV:L (local only) mean an attacker must already own the box at kernel level before this bug becomes relevant, which narrows the real-world urgency compared to a remotely exploitable flaw.

"VBS escape needs SYSTEM first — high-value but deep in the kill chain."
02 · The Attack Path

4 steps from start to impact.

STEP 01

Achieve SYSTEM on the target host

The attacker must first compromise the host to the point of having local administrator or SYSTEM-level code execution. This is a prerequisite, not part of the CVE itself. Common paths include credential theft, lateral movement from another compromised host, or chaining a separate RCE+LPE. The CVSS vector's PR:H confirms this is not an unprivileged escalation.
Conditions required:
  • Local access to the target Windows host
  • SYSTEM or equivalent high-privilege execution context
Where this breaks in practice:
  • Requires a separate initial-access and privilege-escalation chain before this CVE is even reachable
  • EDR solutions should detect most SYSTEM-acquisition techniques (token theft, named-pipe impersonation, service abuse)
Detection/coverage: EDR behavioral detection for privilege escalation to SYSTEM (e.g., CrowdStrike, Defender for Endpoint, SentinelOne)
STEP 02

Identify VBS / Credential Guard status

The attacker verifies that VBS is enabled and the Secure Kernel is loaded. On hosts without VBS, this vulnerability has no meaningful additional impact beyond what SYSTEM already provides. The attacker checks msinfo32, Get-CimInstance Win32_DeviceGuard, or the presence of securekernel.exe in the running module list.
Conditions required:
  • VBS enabled on the target (common on Windows 11 22H2+, Server 2025, and enterprise-managed DCs)
Where this breaks in practice:
  • VBS adoption is growing but not universal — many Windows 10 and older Server deployments do not enable it
  • Without VBS, the attacker already has SYSTEM and gains nothing incremental from this CVE
STEP 03

Trigger the untrusted pointer dereference

The attacker crafts a malicious hypercall or secure system call that passes a pointer value from VTL0 into the Secure Kernel without proper validation. The Secure Kernel dereferences this attacker-controlled pointer, allowing arbitrary read/write within the VTL1 address space. No public PoC exists yet; researcher vmpr00be has announced a forthcoming deep-dive blog with exploit code. The AC:L (low complexity) rating suggests the trigger is straightforward once the prerequisite privilege is met.
Conditions required:
  • Knowledge of the specific Secure Kernel call surface and pointer layout
  • Build-specific offsets (affected: 10.0.28000.0 – 10.0.28000.2953)
Where this breaks in practice:
  • No public exploit code available as of 2026-09-10
  • Secure Kernel attack surface is poorly documented; exploit development requires deep VBS internals expertise
  • HVCI (if co-enabled) constrains the code-execution primitives available in VTL0
Detection/coverage: Kernel ETW telemetry for anomalous secure syscalls; Microsoft Defender for Endpoint may flag VTL boundary violations in future signatures
STEP 04

Extract Credential Guard secrets or disable HVCI

With arbitrary read/write in VTL1, the attacker can dump LSA secrets protected by Credential Guard (NTLM hashes, Kerberos TGTs, DPAPI master keys) or disable HVCI enforcement to load unsigned kernel drivers. On a domain controller, the extracted TGTs enable forging golden tickets or performing pass-the-hash across the domain. On a VBS-protected endpoint, HVCI bypass allows loading a rootkit that persists below EDR visibility.
Conditions required:
  • Successful VTL1 read/write primitive from step 3
Where this breaks in practice:
  • Credential Guard secret extraction requires knowing the in-memory layout of lsaiso.exe within VTL1
  • Domain-controller deployments with Credential Guard are the highest-value targets but represent a subset of all Windows installs
Detection/coverage: Post-exploitation: anomalous Kerberos ticket usage (golden ticket detection), unexpected HVCI policy changes, Credential Guard health monitoring via Windows Event ID 6400-series
03 · Intelligence Metadata

The supporting signals.

In-the-Wild ExploitationNot observed. No CISA KEV listing, no vendor acknowledgment of active exploitation, no GreyNoise or shadow-server activity as of 2026-09-10.
Proof-of-ConceptNot yet public. Researcher vmpr00be has announced a forthcoming technical blog with PoC code. No GitHub repos or exploit-db entries found.
EPSS Score0.00303 (bottom ~15th percentile) — reflects the PR:H + AV:L requirements and absence of public exploit code.
KEV StatusNot listed as of 2026-09-10.
CVSS VectorCVSS:3.1/AV:L/AC:L/PR:H/UI:N/S:C/C:H/I:H/A:H — the S:C (scope change) is the critical element, indicating VTL0→VTL1 boundary escape. Temporal: E:U/RL:O/RC:C (unproven exploit, official remediation available).
Affected VersionsWindows builds 10.0.28000.0 through 10.0.28000.2953 (Windows 11 24H2 / Server 2025 lineage with VBS-capable Secure Kernel).
Fixed Version10.0.28000.2954 — September 2026 cumulative update.
Scanning / ExposureNot remotely exploitable (AV:L), so Shodan/Censys/FOFA exposure data is not applicable. Internal vulnerability scanners (Qualys, Tenable, Rapid7) should flag unpatched builds via authenticated scans.
Disclosure TimelineReserved 2026-08-31, published 2026-09-08 (September Patch Tuesday), advisory updated 2026-09-09.
Reportervmpr00be via coordinated MSRC disclosure. Also reported CVE-2026-69846 and CVE-2026-69906 (same Secure Kernel component).
04 · The Call

noisgate verdict.

Final Verdict
= UNCHANGED to HIGH (7.8/10)

The single most decisive factor sustaining this at HIGH is the scope-change into VTL1 — this is not a redundant kernel LPE but a defeat of the VBS isolation boundary that protects Credential Guard secrets and HVCI policy, making it qualitatively different from a standard SYSTEM escalation. The PR:H prerequisite (attacker must already be SYSTEM) prevents an upgrade to CRITICAL.

HIGH Vulnerability existence and scope-change impact
MEDIUM Exploitability assessment (no public PoC to validate AC:L claim)
LOW Exact population of VBS-enabled hosts in enterprise environments

Why this verdict

  • Scope change is real and consequential. The S:C flag reflects a genuine VTL0→VTL1 escape. On any host running Credential Guard, this converts a kernel compromise into extraction of hypervisor-protected domain credentials — a qualitative escalation that standard LPE bugs do not provide.
  • Role multiplier — domain controllers with Credential Guard. On a DC, VTL1 secrets include cached Kerberos TGTs and NTLM hashes for domain accounts. Extracting these enables golden ticket attacks and domain-wide lateral movement. DCs are a canonical deployment of Credential Guard, meaning the high-value role represents a significant share of the affected component's installed base. Blast radius: host → domain → fleet.
  • PR:H caps severity below CRITICAL. The attacker must already hold SYSTEM-level privileges locally. This is a deep post-exploitation capability, not an initial-access vector. The bug extends an existing compromise rather than enabling a new one, which limits urgency relative to remotely exploitable flaws.
  • No PoC, no exploitation, low EPSS. With EPSS at 0.3%, no public exploit code, and no KEV listing, the near-term weaponization risk is low. However, the announced researcher blog will likely change this calculus — Secure Kernel exploit primitives attract APT-grade interest.

Why not higher?

A CRITICAL rating would require either active exploitation or a lower prerequisite bar. The PR:H requirement means the attacker must already own the kernel — this is not an initial-access or low-privilege escalation vector. Without a public PoC and with zero in-the-wild activity, the immediacy does not justify CRITICAL. If the researcher releases working exploit code and Credential Guard bypass is demonstrated on DCs, reassessment toward CRITICAL would be warranted.

Why not lower?

The VBS boundary escape is not cosmetic — it defeats the *last line of defense* that Credential Guard provides against kernel-level attackers. On domain controllers and high-security endpoints, this is the difference between 'compromised one host' and 'extracted domain-wide secrets that VBS was specifically designed to protect.' The floor for a vulnerability in the Secure Kernel (a hypervisor-tier component by definition) with demonstrated scope change is HIGH.

05 · Compensating Control

What to do — in priority order.

  1. Prioritize patching on DCs and Credential Guard-enabled servers first — These are the hosts where VTL1 escape translates to domain-credential extraction. Apply the September 2026 cumulative update (build 10.0.28000.2954+) to all domain controllers and Tier-0 assets within the noisgate HIGH remediation SLA of 180 days, but front-load DCs to within 30 days.
  2. Audit and restrict local admin / SYSTEM access paths — Since PR:H is the gating prerequisite, reducing the population of accounts and processes that can reach SYSTEM on high-value hosts directly limits exploitability. Review LAPS coverage, remove standing admin rights, enforce just-in-time (JIT) admin via PAM. Deploy within 30 days per noisgate mitigation SLA.
  3. Monitor Credential Guard health events — Windows Event IDs 6400–6499 report Credential Guard operational status. Alert on unexpected Credential Guard failures or policy changes, which could indicate VTL1 tampering. Configure SIEM collection within 30 days.
  4. Enable HVCI alongside Credential Guard — While this CVE can theoretically bypass HVCI via VTL1 write, co-enabling HVCI raises the bar for the VTL0 exploit primitives needed to trigger the pointer dereference. It doesn't block the bug but adds friction to the chain.
  5. Watch for vmpr00be PoC release and re-triage — The researcher has announced a forthcoming technical blog with exploit code. When it drops, reassess urgency — a working PoC for Credential Guard bypass will attract rapid APT adoption. Set a monitoring alert for the researcher's publications.
What doesn't work
  • Network segmentation / firewalls — this is a local-only vulnerability (AV:L). Network controls do not affect the attack path; the attacker is already on the host with SYSTEM.
  • Application allowlisting (AppLocker, WDAC) — the vulnerability is in the Secure Kernel itself, not in a user-mode application. Code integrity policies in VTL0 cannot protect VTL1 from a kernel-level attacker who has already bypassed them.
  • Disabling VBS as a workaround — removing VBS removes the *protection*, not the vulnerability. Without VBS, an attacker with SYSTEM already has unrestricted kernel access; you'd be trading a potential VTL1 escape for guaranteed lack of credential isolation.
06 · Verification

Crowdsourced verification payload.

Run on each target Windows host with administrator privileges. Example: powershell -ExecutionPolicy Bypass -File .\Check-CVE-2026-83939.ps1. The script checks whether the installed OS build is in the affected range and whether VBS/Credential Guard is active (which determines real-world exposure).

noisgate-verify.ps1
POWERSHELLREAD-ONLYSAFE
#Requires -RunAsAdministrator
# Check-CVE-2026-83939.ps1
# Checks for CVE-2026-83939: Untrusted pointer dereference in Windows Secure Kernel Mode
# Output: VULNERABLE / PATCHED / UNKNOWN
# Exit codes: 1 = VULNERABLE, 0 = PATCHED, 2 = UNKNOWN

$ErrorActionPreference = 'Stop'

try {
    $build = [System.Environment]::OSVersion.Version
    $fullBuild = (Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion' -Name UBR -ErrorAction SilentlyContinue).UBR
    $buildStr = "$($build.Major).$($build.Minor).$($build.Build).$fullBuild"
    Write-Host "[*] OS Build: $buildStr"

    # Check if this is the affected build family (28000.x)
    if ($build.Build -ne 28000) {
        Write-Host "[*] Build family $($build.Build) is outside the known affected range (28000)."
        Write-Host "UNKNOWN"
        exit 2
    }

    # Affected: 28000.0 through 28000.2953, fixed at 28000.2954+
    if ($null -eq $fullBuild) {
        Write-Host "[!] Could not determine UBR (Update Build Revision)."
        Write-Host "UNKNOWN"
        exit 2
    }

    # Check VBS / Credential Guard status for exposure context
    $dg = Get-CimInstance -ClassName Win32_DeviceGuard -Namespace 'root\Microsoft\Windows\DeviceGuard' -ErrorAction SilentlyContinue
    $vbsRunning = ($dg.VirtualizationBasedSecurityStatus -eq 2)
    $credGuard = ($dg.SecurityServicesRunning -contains 1)
    Write-Host "[*] VBS Running: $vbsRunning | Credential Guard Active: $credGuard"

    if ($fullBuild -le 2953) {
        Write-Host "[!] Build $buildStr is in the AFFECTED range (fixed in 28000.2954)."
        if ($vbsRunning) {
            Write-Host "[!] VBS is ACTIVE - this host is exploitable if an attacker achieves SYSTEM."
        } else {
            Write-Host "[*] VBS is NOT active - vulnerability exists but practical impact is reduced."
        }
        Write-Host "VULNERABLE"
        exit 1
    } else {
        Write-Host "[+] Build $buildStr is at or above the fix (28000.2954)."
        Write-Host "PATCHED"
        exit 0
    }
} catch {
    Write-Host "[!] Error during check: $_"
    Write-Host "UNKNOWN"
    exit 2
}
07 · Bottom Line

If you remember one thing.

TL;DR
Monday morning: Identify all Windows build 28000.x hosts via your vulnerability scanner or SCCM/Intune and cross-reference against Credential Guard / VBS enablement status. Prioritize the September 2026 cumulative update (build 28000.2954) on domain controllers and Tier-0 servers running Credential Guard — these are where this bug translates to domain-credential extraction. Per the noisgate mitigation SLA for HIGH, deploy compensating controls (restrict SYSTEM-access paths, enable Credential Guard health monitoring) within 30 days. Per the noisgate remediation SLA, patch all affected hosts within 180 days, but front-load DCs and identity-tier servers into your first patch wave. No immediate emergency action is required — there is no public PoC and no active exploitation — but set a watch for researcher vmpr00be's forthcoming technical blog, which will include exploit code and could shift this to CRITICAL overnight.

Sources

  1. BrinzTech — Three Elevation of Privilege Vulnerabilities in Windows Secure Kernel
  2. THREATINT CVE-2026-83939 Entry
  3. CrowdStrike — September 2026 Patch Tuesday Analysis
  4. Qualys — Microsoft Patch Tuesday September 2026 Review
  5. Cyber Security News — September 2026 Patch Tuesday (973 CVEs)
  6. ParagMali — Windows Secure Kernel Architecture Deep Dive
  7. Talos — Microsoft Patch Tuesday September 2026
Peer Review

What defenders are saying.

Submit a review attribution: handle + country only
0 flags selected · stored anonymously
Validation Results

Crowdsourced verification outputs.

Results submitted by users who ran the verification payload against their environment.