Someone slipped a timing bomb into the service that shares a process with your Active Directory identity plane
CVE-2026-62820 is a race-condition flaw (CWE-362) in the Windows DNS Server service that allows an unauthenticated remote attacker to achieve code execution by sending specially crafted packets over the network. No user interaction is required. The vulnerability affects Windows Server 2022 and Windows Server 2025 (both Server Core and full installations) running the DNS Server role, and was disclosed as part of the August 11, 2026 Patch Tuesday release alongside several related DNS Server RCEs (CVE-2026-62878, CVE-2026-62817, CVE-2026-65789). Patches are available via KB5120233 and KB5120228 (hotpatch).
Microsoft rates this HIGH at 8.1 with AC:H to account for the race-condition requirement. That score is technically fair in isolation but dangerously misleading in context. The CVSS vector does not encode *where* the DNS Server role actually runs. In the vast majority of Active Directory environments, Windows DNS is co-located on domain controllers — the single most privileged tier in the enterprise. SYSTEM-level code execution on a DC is not a server compromise; it is a domain compromise. The vendor score treats this as 'one server goes down,' but reality says 'your entire identity plane is owned.' noisgate upgrades to CRITICAL.
4 steps from start to impact.
Identify target DNS server
- Network reachability to port 53 on target
- Internal-only DNS servers require the attacker to already have internal network access
Send crafted DNS packets to trigger race
- Ability to send UDP/TCP packets to port 53
- No authentication required
- Race condition is non-deterministic — individual attempts may fail
- Packet filtering or rate limiting on DNS could slow spraying
- No public PoC exists yet, so the attacker needs to develop the exploit independently
Win the race — achieve code execution as SYSTEM
- Successful race-condition win
- Non-deterministic success — may require sustained packet spray over minutes
- Environmental factors (CPU load, network latency) affect race timing
Pivot from DC SYSTEM to domain compromise
lsadump::dcsync), dump NTDS.dit, create golden tickets, or add themselves to Domain Admins. This is the standard post-exploitation path from DC compromise to full AD takeover. The entire forest is now controlled.- Code execution achieved on a domain controller running DNS
- Only applies when DNS role is on a DC (but this is the dominant deployment pattern — estimated >70% of Windows DNS installs)
The supporting signals.
| In-the-wild exploitation | Not observed. Not on CISA KEV. Microsoft marks exploitation as "more likely", meaning they assess the bug is reliably weaponizable. |
|---|---|
| Proof-of-concept | No public PoC as of 2026-08-13. No known repos on GitHub or exploit-db. The race-condition requirement raises the bar for independent development, but related DNS Server RCE PoCs (CVE-2026-62878) may accelerate research. |
| EPSS | 0.00453 (≈ top 5%). Low but non-trivial for a 2-day-old CVE — expect this to climb as researchers analyze the patch diff. |
| KEV status | Not listed as of 2026-08-13. |
| CVSS vector | CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H — Network attack vector, high complexity (race condition), no privileges, no interaction, full CIA impact. The AC:H is doing all the work keeping this below 9.8. |
| Affected versions | Windows Server 2022, Windows Server 2025 (including Server Core). Likely affects Server 2019 as well per the DNS Server codebase lineage, pending full MSRC matrix confirmation. |
| Fixed versions | August 2026 cumulative updates: KB5120233 (Windows Server 2025), KB5120228 (Server 2025 hotpatch). Check MSRC for Server 2022/2019 specific KBs. |
| Scanning/exposure | DNS (port 53) is universally exposed internally. Shodan indexes ~2.8 million Windows DNS servers externally. GreyNoise has not yet tagged specific CVE-2026-62820 scanning activity. |
| Disclosure date | 2026-08-11 (Microsoft Patch Tuesday) |
| Related CVEs | CVE-2026-62878 (DNS Server wormable RCE, CVSS 9.8), CVE-2026-62817 (DNS Server RCE, 8.8), CVE-2026-65789 (DNS Server RCE, 8.1) — treat as a cluster. |
noisgate verdict.
The single most decisive factor is the role multiplier: Windows DNS Server is canonically co-located on Active Directory domain controllers, meaning SYSTEM-level RCE on this service is functionally equivalent to domain takeover in >70% of deployments. While the race condition reduces exploit reliability per-attempt, the unauth/no-interaction attack surface on a high-throughput service allows sustained spraying that erodes this friction over time.
Why this verdict
- Role multiplier: Windows DNS Server runs on domain controllers in the vast majority of Active Directory deployments (estimated >70% of installs). Successful RCE grants SYSTEM on a DC → DCSync → domain admin → fleet compromise. This is not a single-server blast radius — it is identity-plane-scale. Per the noisgate role framework, this floors the verdict at CRITICAL.
- Unauthenticated, no-interaction, network-reachable: The CVSS vector is
AV:N/PR:N/UI:N, meaning any host that can reach port 53 can attempt exploitation. DNS is one of the most universally exposed services in any network. There is zero attacker prerequisite beyond network access. - Race condition friction is real but erodible:
AC:Hreflects that individual attempts are non-deterministic. However, DNS processes thousands of concurrent queries — an attacker can spray crafted packets at high volume to dramatically increase the probability of winning the race. This is not equivalent to the friction of, say, requiring a specific non-default configuration. - Microsoft's own assessment says 'exploitation more likely': This is Microsoft's signal that the race window is practically winnable, not just theoretically possible. Combined with Talos already shipping Snort detection rules, the security community is treating this as imminently weaponizable.
Why not higher?
This is already assessed at CRITICAL. A score above 8.5 (toward 9.0+) would be warranted if a public PoC existed, if exploitation were observed in the wild, or if the race condition were trivially winnable on first attempt. None of those conditions are met today.
Why not lower?
Downgrading to HIGH would require evidence that fewer than 10% of Windows DNS Server installs occupy the DC role — the opposite is true. The race condition alone is insufficient to break below the CRITICAL floor because DNS is a high-throughput service where spray attacks erode timing-based friction, and Microsoft themselves rate exploitation as 'more likely.' The unauth/no-interaction attack surface eliminates the friction points (phishing, credential theft, VPN access) that justify downgrades in other CVEs.
What to do — in priority order.
- Restrict DNS Server access via Windows Firewall or network segmentation — Limit which hosts and subnets can reach port 53 on your DCs. Most clients only need recursive resolution from their local site DC. Block external inbound DNS to internal DCs entirely. Deploy within 3 days per the noisgate CRITICAL mitigation SLA.
- Deploy IDS/IPS rules for CVE-2026-62820 — Talos has published Snort rules (SIDs 66902–66948) covering this CVE family. Deploy on inline IPS sensors protecting DNS server segments. This gives you a detection/blocking layer while patching proceeds.
- Enable dns.exe process monitoring via EDR — Create high-fidelity alerts for any child process spawned by
dns.exeon domain controllers. SYSTEM-level code execution from the DNS service will manifest as anomalous process creation. Most EDR platforms can deploy this rule within hours. - Separate DNS role from DC role where architecturally feasible — For new deployments or during upcoming maintenance windows, consider running DNS on dedicated member servers rather than DCs. This eliminates the DC blast radius for DNS-targeting exploits. This is a longer-term architectural change.
- Prioritize patching DCs running DNS first — If you cannot patch all DNS servers within 3 days, triage by role: domain controllers running DNS are the highest-value targets. Patch these first, then standalone DNS servers.
- DNS-over-HTTPS (DoH) or DNS-over-TLS (DoT) on clients — these protect client-to-resolver privacy but do not change how the server processes incoming queries internally. The race condition is in server-side packet handling, not the transport layer.
- Response Rate Limiting (RRL) — RRL throttles *outbound* DNS responses to mitigate reflection attacks. It does not limit *inbound* query processing where the race condition lives.
- Disabling DNS recursion — authoritative-only servers are still vulnerable because the flaw is in core DNS packet processing, not in the recursion logic specifically.
Crowdsourced verification payload.
Run this on each Windows Server host that has the DNS Server role installed. Execute as Administrator in an elevated PowerShell session. Example: .\Check-CVE-2026-62820.ps1
#Requires -RunAsAdministrator
<#
.SYNOPSIS
Checks whether the host is patched against CVE-2026-62820 (Windows DNS Server RCE).
.DESCRIPTION
Verifies the DNS Server role is installed, then checks for the August 2026
cumulative update (KB5120233 or KB5120228) via hotfix history.
Outputs VULNERABLE, PATCHED, or UNKNOWN.
.NOTES
Must run as Administrator on the target DNS server.
#>
$ErrorActionPreference = 'Stop'
# Step 1: Check if DNS Server role is installed
try {
$dnsFeature = Get-WindowsFeature -Name DNS -ErrorAction SilentlyContinue
if (-not $dnsFeature -or -not $dnsFeature.Installed) {
Write-Host 'UNKNOWN - DNS Server role is not installed on this host. CVE-2026-62820 does not apply.'
exit 2
}
} catch {
# Fallback for systems without Get-WindowsFeature
$dnsSvc = Get-Service -Name DNS -ErrorAction SilentlyContinue
if (-not $dnsSvc) {
Write-Host 'UNKNOWN - DNS Server service not found. CVE-2026-62820 does not apply.'
exit 2
}
}
# Step 2: Check for August 2026 patches
$targetKBs = @('KB5120233', 'KB5120228')
$installedHotfixes = Get-HotFix -ErrorAction SilentlyContinue | Select-Object -ExpandProperty HotFixID
$patched = $false
foreach ($kb in $targetKBs) {
if ($installedHotfixes -contains $kb) {
$patched = $true
Write-Host "PATCHED - $kb is installed. CVE-2026-62820 is remediated."
exit 0
}
}
# Step 3: Secondary check - OS build number (Server 2025)
$osBuild = [System.Environment]::OSVersion.Version.Build
$osRevision = (Get-ItemProperty 'HKLM:\SOFTWARE\Microsoft\Windows NT\CurrentVersion' -Name UBR -ErrorAction SilentlyContinue).UBR
if (-not $patched) {
# Check if any KB installed after August 11, 2026 (patch date)
$augustPatches = Get-HotFix -ErrorAction SilentlyContinue | Where-Object {
$_.InstalledOn -ge [datetime]'2026-08-11'
}
if ($augustPatches) {
Write-Host "UNKNOWN - Post-August-11 hotfix found ($($augustPatches[0].HotFixID)) but not the specific KB for CVE-2026-62820. Verify manually."
exit 2
}
}
Write-Host "VULNERABLE - DNS Server role is installed and August 2026 security update is MISSING. CVE-2026-62820 is exploitable."
exit 1If you remember one thing.
Sources
- Microsoft MSRC Advisory – CVE-2026-62820
- CrowdStrike – August 2026 Patch Tuesday Analysis
- Cisco Talos – August 2026 Patch Tuesday Snort Rules
- Tenable – August 2026 Patch Tuesday Overview
- Security Affairs – August 2026 Patch Tuesday: Zero-Day and Wormable RCE
- Rapid7 – Patch Tuesday August 2026
- Tenable Nessus Plugin – KB5120233 Windows Server 2025
What defenders are saying.
Crowdsourced verification outputs.
Results submitted by users who ran the verification payload against their environment.