Someone hands a poisoned letter to your castle's gatekeeper — and now they own the drawbridge
CVE-2026-62817 is an out-of-bounds write (CWE-787) in the Windows DNS Server service. An unauthenticated attacker on the same Layer 2 network segment can send specially crafted DNS packets to trigger a memory corruption condition, achieving remote code execution as SYSTEM. The flaw affects Windows Server 2019, 2022, and 2025 — plus, notably, Windows 10 (all supported versions through 22H2) and Windows 11 (through 26H1) when the DNS Server role is enabled. It was disclosed on 2026-08-11 as part of Microsoft's August 2026 Patch Tuesday, alongside three sibling DNS Server RCE bugs: CVE-2026-62878 (stack overflow, CVSS 9.8), CVE-2026-62820 (race condition, 8.1), and CVE-2026-65789 (use-after-free, 8.1).
Microsoft rates this Important / HIGH at 8.8, and the score is honest given the CVSS vector's AV:A constraint — the attack is limited to an adjacent network, not the internet. However, the vendor score doesn't tell you *where* the DNS Server role actually runs in production. In the overwhelming majority of Active Directory environments, DNS Server is co-located on domain controllers. That means a successful exploit doesn't just give you SYSTEM on 'a DNS server' — it gives you SYSTEM on a domain controller, which is one hop from DA and full domain compromise. The vendor score is mechanically correct but understates the blast radius for canonical AD deployments.
4 steps from start to impact.
Attain Layer 2 adjacency to DNS server
- Physical or logical Layer 2 adjacency to the DNS server's network segment
- Properly segmented networks place DCs on isolated VLANs with ACLs restricting L2 access
- 802.1X or NAC on server VLANs blocks rogue devices
- Many enterprises use inter-VLAN routing rather than true L2 adjacency, which may not satisfy AV:A
Send crafted DNS packets to trigger OOB write
PR:N) and no user interaction is needed (UI:N). The malformed packet triggers the out-of-bounds write in the DNS Server service's parsing logic, corrupting heap or stack memory depending on the specific write primitive.- Network reachability to UDP/53 or TCP/53 on the target
- Target running the Windows DNS Server role with the vulnerable binary
- No public proof-of-concept exploit is available as of 2026-08-13
- Crafting a reliable OOB write exploit requires bypassing ASLR and DEP on modern Windows Server
- Windows Defender Exploit Guard / CFG may complicate exploitation
Achieve SYSTEM-level code execution on DNS Server
NT AUTHORITY\SYSTEM. The attacker now has full control of the host — file system, registry, memory, and credential stores.- Successful memory corruption and control-flow hijack from step 2
- Kernel-mode protections (Credential Guard, HVCI) on Server 2022/2025 limit post-exploitation credential theft
- EDR agents should detect anomalous child processes spawned from
dns.exe
Pivot from DNS Server to domain compromise
lsadump::dcsync), dump NTDS.dit, extract the KRBTGT hash, and forge Golden Tickets. From a single DC, the entire AD forest is compromised within minutes.- Target host is a domain controller (true for >80% of Windows DNS Server deployments)
- Credential Guard not enforced on the DC
- Environments running Credential Guard on DCs significantly limit credential extraction
- Tiered administration models (PAW/T0 isolation) limit lateral spread from DC compromise
- AD monitoring tools (Microsoft Defender for Identity, Semperis DSP) detect DCSync and Golden Ticket activity
The supporting signals.
| In-the-wild exploitation | None observed. Microsoft confirms neither exploited nor publicly disclosed as of 2026-08-12. No CISA KEV listing. |
|---|---|
| Proof-of-concept | No public PoC available as of 2026-08-13. No named researcher PoC repos found on GitHub. Given the AV:A constraint and memory corruption complexity, weaponization timeline is likely weeks to months. |
| EPSS | 0.0057 (0.57%) — bottom quartile. The adjacent-network requirement and lack of PoC keep exploitation probability low in the near term. |
| KEV status | Not listed. No CISA KEV entry as of 2026-08-13. |
| CVSS vector | CVSS:3.1/AV:A/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H — Adjacent network, low complexity, no privileges, no interaction, unchanged scope, full CIA impact. The AV:A is the key limiter vs. the sibling CVE-2026-62878 which is AV:N / 9.8. |
| Affected versions | Windows Server 2019, 2022, 2025 (DNS Server role). Also affects Windows 10 (1809, 21H2, 22H2) and Windows 11 (23H2, 24H2, 25H2, 26H1) when DNS Server role is installed — rare on client SKUs. |
| Fixed versions | August 2026 cumulative updates (released 2026-08-12). Specific KB numbers vary by OS version — consult Microsoft Security Update Guide for exact KBs per Server edition. |
| Scanning/exposure data | Shodan shows ~2.1M Windows DNS servers globally, but AV:A means internet-facing exposure is irrelevant — the attack requires Layer 2 adjacency, not IP reachability. The real exposure metric is internal network segmentation posture. |
| Disclosure date | 2026-08-11 (August 2026 Patch Tuesday) |
| Related CVEs | Part of a cluster of four DNS Server RCEs: CVE-2026-62878 (9.8, stack overflow, AV:N), CVE-2026-62820 (8.1, race condition), CVE-2026-65789 (8.1, UAF). Prioritize CVE-2026-62878 first as it is network-exploitable. |
noisgate verdict.
The verdict remains HIGH because the DNS Server role's canonical deployment on Active Directory domain controllers creates a blast radius of full domain compromise upon successful exploitation. The single most decisive factor preventing an upgrade to CRITICAL is the AV:A (adjacent network) constraint — the attacker must already hold Layer 2 adjacency to the DC's VLAN, which in segmented environments represents a meaningful post-compromise prerequisite that compounds with the absence of any public PoC or in-the-wild exploitation.
Why this verdict
- AV:A constraint already priced in: The vendor's 8.8 CVSS already reflects the adjacent-network limitation. The score is mechanically correct. We shave 0.8 points for the additional real-world friction of VLAN segmentation in mature enterprises, no PoC availability, and zero in-the-wild exploitation.
- Role multiplier: DNS Server ≈ Domain Controller. In >80% of AD deployments, the Windows DNS Server role runs directly on domain controllers. RCE as SYSTEM on a DC yields DCSync → KRBTGT extraction → Golden Ticket → full domain/forest compromise. This chain succeeds trivially once step 3 is reached. Blast radius = domain-scale to forest-scale. This floors the verdict at HIGH regardless of other friction.
- No exploitation signal suppresses urgency but not severity. Zero EPSS percentile, no KEV, no PoC, no campaign attribution. This is a theoretical risk today — but DNS parsing bugs have historically been weaponized within 30-60 days of patch release (cf. SIGRed / CVE-2020-1350). The clock is ticking.
- Sibling CVE-2026-62878 (9.8, AV:N) overshadows but does not eliminate this risk. Defenders will rightly prioritize the network-exploitable sibling, but CVE-2026-62817 remains a potent post-compromise escalation vector for attackers already inside the network perimeter.
Why not higher?
Upgrading to CRITICAL would require either active exploitation evidence or a network-reachable (AV:N) attack vector. The AV:A constraint means this bug cannot be exploited from the internet or even from a different VLAN without prior compromise of the DC's network segment. No PoC exists, EPSS is sub-1%, and there is no KEV listing. The sibling CVE-2026-62878 at 9.8 with AV:N is the one that warrants CRITICAL treatment.
Why not lower?
Downgrading to MEDIUM would ignore that the canonical target — a domain controller — represents one of the highest-value hosts in any enterprise. An unauthenticated, low-complexity RCE on a DC is a domain-ending event. The AV:A requirement is real friction but does not eliminate the threat: flat networks, misconfigured VLANs, and compromised hosts on server segments are common realities. The DC role multiplier floors this at HIGH.
What to do — in priority order.
- Isolate DCs on dedicated VLANs with strict L2 ACLs — The
AV:Aattack vector requires Layer 2 adjacency. Placing domain controllers on a dedicated VLAN with switch-level ACLs that deny all non-essential L2 traffic from workstation/user segments is the single most effective compensating control. Validate or deploy within the noisgate mitigation SLA of 30 days. Most mature environments already have this; verify it's enforced, not just documented. - Deploy IDS/IPS signatures for malformed DNS traffic — Cisco Talos released Snort rules for the August 2026 Patch Tuesday DNS cluster. Deploy these on network segments where DNS servers reside. This provides detection-in-depth even if segmentation has gaps. Deploy within 30 days per noisgate mitigation SLA.
- Enable Windows Defender Exploit Guard on DNS servers — Configure Exploit Guard's mandatory ASLR, CFG, and export address filtering for
dns.exe. This raises the exploitation bar significantly for memory corruption bugs. Apply via Group Policy to all servers running the DNS role. - Monitor dns.exe for anomalous behavior via EDR — Create detection rules for
dns.exespawning unexpected child processes (cmd.exe, powershell.exe, rundll32.exe). Any code execution from the DNS service should be treated as a critical alert. Most EDR platforms can scope this within hours. - Audit DNS Server role placement — Identify any DNS Server instances running on non-DC hosts (uncommon but possible). If DNS is running on member servers or client SKUs, evaluate whether the role can be removed. Reducing the attack surface is preferable to defending it.
- DNS-over-HTTPS (DoH) or DNS-over-TLS (DoT) at the client — These encrypt DNS traffic in transit but do not change how the server parses incoming queries. The OOB write is in the server-side parsing logic and is triggered regardless of transport encryption.
- Windows Firewall rules blocking external DNS — The attack vector is adjacent-network, not internet-facing. Blocking inbound DNS from the internet is good hygiene but does not mitigate this CVE, which is exploited from inside the network.
- Disabling DNS recursion — The vulnerability is in the DNS Server parsing code, not in recursive resolution logic. Disabling recursion does not prevent a crafted query from reaching the vulnerable code path.
Crowdsourced verification payload.
Run this script on each Windows Server host running the DNS Server role (typically your domain controllers). Execute as Administrator in an elevated PowerShell session. Example: .\Check-CVE-2026-62817.ps1. No external dependencies required.
#Requires -RunAsAdministrator
<#
.SYNOPSIS
Checks whether the Windows DNS Server role is installed and whether the
August 2026 cumulative update has been applied (CVE-2026-62817 mitigation).
.NOTES
Outputs VULNERABLE, PATCHED, or UNKNOWN.
Exit codes: 1 = VULNERABLE, 0 = PATCHED, 2 = UNKNOWN
#>
$ErrorActionPreference = 'Stop'
# Check if DNS Server role is installed
$dnsRole = Get-WindowsFeature -Name DNS -ErrorAction SilentlyContinue
if (-not $dnsRole -or -not $dnsRole.Installed) {
Write-Host "UNKNOWN - DNS Server role is not installed on this host. Not affected."
exit 2
}
Write-Host "[*] DNS Server role is INSTALLED on this host."
# Check for August 2026 cumulative updates
# August 2026 Patch Tuesday was 2026-08-11; updates released 2026-08-12
$augustPatches = Get-HotFix | Where-Object {
$_.InstalledOn -ge [DateTime]'2026-08-11'
} | Sort-Object InstalledOn -Descending
if ($augustPatches.Count -gt 0) {
$latest = $augustPatches[0]
Write-Host "PATCHED - Found post-August-2026 update: $($latest.HotFixID) installed on $($latest.InstalledOn.ToString('yyyy-MM-dd'))."
Write-Host "Verify this KB against Microsoft Security Update Guide for CVE-2026-62817 coverage."
exit 0
} else {
# Cross-check dns.exe file version as secondary signal
$dnsExe = "$env:SystemRoot\System32\dns.exe"
if (Test-Path $dnsExe) {
$ver = (Get-Item $dnsExe).VersionInfo.ProductVersion
Write-Host "[*] dns.exe version: $ver"
}
Write-Host "VULNERABLE - No cumulative update from August 2026 or later detected."
Write-Host "Apply the August 2026 cumulative update immediately."
exit 1
}If you remember one thing.
Sources
What defenders are saying.
Crowdsourced verification outputs.
Results submitted by users who ran the verification payload against their environment.