WarrenBarr
SEC-01
Advisories
Published findings
OPS-02
Methodology
How an engagement runs
ENG-03
IT & Dev
Build and administer
ARM-04
Software
Tooling I wrote
MSP-05
MSSP
Flat-rate managed
REC-06
Case Studies
Work, in full
WEB-07
Websites
Sites that convert
SUP-08
Gear
Coming soon
DOC-09
Doctrine
What I will not do
WHO-10
About
Who you'd be hiring

N-able N-central Authentication Bypass — CVE-2026-18556 / CVE-2026-18577
The First Patch Did Not Hold · Attackers Pivoted Into Managed Endpoints · CISA KEV

CVE-2026-18556 CVE-2026-18577 CWE-288 CVSS 8.2 T1199 T1572 RMM CISA KEV Actively Exploited
N-able N-central is the remote monitoring and management platform a great many managed service providers use to administer their customers' endpoints. Two authentication bypasses in it — CVE-2026-18556 and CVE-2026-18577, both CVSS 8.2, both classed CWE-288 (authentication bypass using an alternate path) — allow an unauthenticated remote attacker to take over the N-central server as an administrator.
That is not a server compromise. That is a compromise of every endpoint that server manages.
The first patch was incomplete. N-able fixed one exploitation path in 2026.2, then discovered attackers reaching the same vulnerability by another route. Only build 2026.3.1.7, shipped 2 August 2026, is unaffected. If you upgraded to 2026.3 on the original instruction, you are still vulnerable.
N-able began investigating on 31 July 2026 — not because of a security alert, but because on-premises customers were generating an unusual volume of licensing errors. The investigation found an attacker had already gained remote administrative access to servers running 2026.1 and earlier.
That detail is worth sitting with. The signal that surfaced a live intrusion into an MSP control plane was a billing subsystem behaving oddly. Nothing in the security stack raised its hand first.
1. Unauthenticated request -> admin access on the N-central server 2. Built-in "Take Control" -> reach into MANAGED CUSTOMER ENDPOINTS 3. On each endpoint -> register a Cloudflare tunnel AS A SERVICE 4. Tunnel dials OUT to Cloudflare's edge - no inbound firewall rule needed - no listening port to find - runs as a service, so it survives reboot 5. Access persists AFTER the N-central route is revoked
Step four is the part that makes this expensive. An outbound tunnel to a reputable CDN looks like ordinary traffic, needs nothing opened at the perimeter, and does not appear in a port scan. Step five is the part that makes it dangerous: patching N-central does not evict the attacker, because the persistence lives on somebody else's machine.
Nothing here suggests Cloudflare was compromised. Its tunneling service was abused, the same way any legitimate outbound-connect tool can be.
Huntress reported exploitation in one partner account running a self-hosted instance. From it the attackers reached nine organisations, one endpoint in each. On the evidence available, post-compromise activity was limited to enumerating running processes before disconnecting — and Huntress did not observe the Cloudflare tunnel installation N-able described.
Read that as reconnaissance, not restraint. Somebody walked the building and left. N-able has not disclosed how many customers were affected, how many downstream devices were reached, when exploitation began, who is responsible, or whether data was taken.
ATTACKER IPs (published by N-able) 173.249.252.200 87.249.138.34 37.19.210.32 37.153.90.88 92.118.112.181 68.235.46.214 Huntress identified four of these as Mullvad or NordVPN exit nodes. Treat a match as a lead to correlate, not as proof on its own. DOMAINS (Huntress) mousears.synology.me wagoosh.direct.quickconnect.to who-ripped-one.direct.quickconnect.to (Synology QuickConnect infrastructure) ON-HOST svchost.exe located in a user's Documents folder a service named "Cloudflared" outbound tunnel traffic from endpoints that have no business making it LOGS TO PULL ui_access_control.log (N-central) C:\ProgramData\GetSupportService_N-Central\Logs\BASupSrvc_*.log.gz Sessions attributed to support identities e.g. mspsupport@n-able.com
A caution on those logs. BASupSrvc and ui_access_control entries are also written during legitimate Take Control sessions. Their presence is not evidence of compromise. What matters is a session you cannot tie to a technician, a ticket and a time.
# A tunnel registered as a service is the tell Get-Service | Where-Object { $_.Name -match 'cloudflared|argo|tunnel' } | Select-Object Name, DisplayName, Status, StartType Get-CimInstance Win32_Service | Where-Object { $_.PathName -match 'cloudflared|\.exe.*tunnel.*run' } | Select-Object Name, PathName, StartMode, State # A system binary name in a place it never belongs Get-ChildItem "$env:USERPROFILE\Documents","C:\Users\*\Documents" -Recurse ` -Include svchost.exe,rundll32.exe,services.exe -ErrorAction SilentlyContinue | Select-Object FullName, Length, CreationTime
// Defender / Sentinel — a service binary running from a user profile DeviceProcessEvents | where FileName =~ "cloudflared.exe" or ProcessCommandLine has_all ("tunnel","run") or (FileName =~ "svchost.exe" and FolderPath has_any ("\\Documents\\","\\Downloads\\")) | project Timestamp, DeviceName, AccountName, FileName, FolderPath, ProcessCommandLine, InitiatingProcessFileName

Get to 2026.3.1.7. Not 2026.3 — that was the original guidance and it is no longer sufficient. Hosted NCOD instances are being upgraded automatically on a schedule N-able communicates to partners; self-hosted servers are the customer's job, which in practice means yours.

Then assume the endpoints, not just the server. This is the step people will skip. Upgrading N-central removes the attacker's route in; it does nothing about a tunnel service already installed on a managed machine. Every endpoint reachable from a compromised server has to be checked for the indicators above.

Audit Take Control history. Pull ui_access_control.log and reconcile every session against a real technician and a real ticket. Any session you cannot account for is an investigation, particularly ones attributed to vendor support identities.

Rotate what the server could reach. Administrative credentials, agent registration tokens, API keys, and any credential stored in or accessible from N-central. Administrative access to an RMM is administrative access to its secrets.

Egress-filter your endpoints. This chain depends entirely on an outbound connection to a CDN edge. Endpoints that only ever need to reach a defined set of destinations should not be able to open arbitrary tunnels, and a default-deny egress policy would have stopped step four cold.

The RMM is the highest-value target in any managed environment, because it is the one system deliberately given administrative reach into every other one. That is not a flaw in the model — it is what the model is for. It does mean the control plane deserves the paranoia normally reserved for domain controllers: restricted management interface, MFA that cannot be bypassed, network segmentation, and an alert on every administrative session that nobody can explain.
The incomplete-patch story is the more uncomfortable lesson. A vendor fixed the path they found, shipped it, and attackers found another way to the same door. If your patch management stops at "we applied the update" without tracking whether the fix was later superseded, this is exactly the shape of the gap you will fall into.
FCEB agencies were directed to remediate by 6 August 2026. Finland's national cyber security centre advised that every version available before the emergency hotfix should be treated as vulnerable.
N-able — N-central 2026.3 Hotfix 1 mitigation notice, 2 August 2026
CISA KEV — Known Exploited Vulnerabilities Catalog (both CVEs added August 2026)
Huntress — N-able vulnerability exploitation rapid response, 3 August 2026
Finland NCSC-FI — Vulnerability advisory 2026-21, 2 August 2026
More like this, when it happens Advisories only. No newsletter, no marketing, unsubscribe in one click.

Barr Cyber — Developer Endpoint & Supply Chain Hardening

Developer workstations hold cloud credentials, signing keys and live sessions — which is exactly why this campaign targets them. If your team installs open-source tooling or AI capabilities as part of normal work, and nothing is watching what gets executed from AppData, that gap is worth closing before it is found for you.

Get in Touch →