N-able confirmed on August 2, 2026 that attackers exploited an authentication bypass in its N-central platform to seize administrator accounts and plant persistence on managed endpoints. Its earlier fix for the same flaw proved incomplete.
Quick Summary – TLDR:
- N-able shipped build 2026.3.1.7 on August 2. Every earlier build is exposed to CVE-2026-18577.
- Attackers used the Take Control feature to reach managed endpoints, then registered Cloudflare tunnels as services to keep access.
- Huntress found 55.6% of reachable N-central cloud servers in its telemetry were still unpatched when it published.
- Upgrading the N-central server does not remove tunnel services already installed on managed machines.
- Both CVEs score 8.2 on CVSS 4.0, and N-able assigned each of them itself.
What Happened?
N-able opened an investigation on July 31, 2026 after an unusual volume of licensing errors arrived from on-premises N-central customers. The company found that an attacker had remotely gained administrative access to servers running version 2026.1 and earlier, and it says it identified and contacted a limited number of affected customers without giving a figure.
The attackers then abused Take Control, the platform’s built-in remote session feature, to reach the endpoints those servers manage. “Once on those devices, the attackers registered a new service for a CloudFlare tunnel, enabling persistence into an environment after access to the N-central server was revoked,” N-able said in its security update. Cloudflare tunnels dial outbound, so they need no inbound firewall rule and no open listening port. Nothing in the disclosure suggests Cloudflare itself was compromised.
Hosted NCOD instances are upgraded automatically on a schedule N-able is giving partners. Self-hosted servers must be patched by the customer, and the hotfix notice says agents need no immediate update.
Critical N-able N-central Vulnerability and Active Exploitation: Update.
— Huntress (@HuntressLabs) August 3, 2026
Read more: https://t.co/Vx0LNASOYK pic.twitter.com/nhdpT7yNnI
The First Fix Left the Same Door Open
The original flaw, CVE-2026-18556, is titled “unauthenticated administrative account takeover” in N-able’s own CVE record and classified as CWE-288, an authentication bypass through an alternate path or channel. It covered releases through 2026.1 and was fixed in 2026.2.
Analysis on the morning of August 2 turned up a second route to the same weakness that the first patch did not block. That became CVE-2026-18577 and pushed the affected range out to every build before 2026.3.1.7. Finland’s National Cyber Security Centre said all versions released before the emergency hotfix were vulnerable. Neither CVE record names the vulnerable endpoint or the request sequence.
Last year CISA added two exploited N-central flaws to its Known Exploited Vulnerabilities catalog and gave agencies seven days to patch. Those two required authentication. This one does not.
Huntress Puts Numbers on the Exposure
Huntress published a rapid response on August 3 describing a self-hosted N-central instance inside a single partner account. The attackers reached nine organizations under that account and touched one endpoint in each. On the evidence so far, the post-compromise activity stopped at enumerating running processes before the attackers disconnected, and Huntress did not observe the Cloudflare installation N-able described.
The wider number is the uncomfortable one. “At the time of posting this update, more than half (55.6%) of our partners’ and customers’ reachable cloud servers were still unpatched. That is especially concerning because the N-able server runs a custom distribution of AlmaLinux 9, and does not often have EDR software deployed on it due to running as an appliance,” Huntress researchers wrote. Huntress also identified four of the addresses on N-able’s initial indicator list as Mullvad or NordVPN exit nodes, which limits their attribution value.
An unmonitored appliance is where detection gaps get expensive. Average breach detection time runs 207 days, and United States breaches average $10.22 million against a $4.88 million global figure.
What Defenders Should Check Today?
Patching the server closes the entry route. It does nothing about a tunnel service already sitting on a client workstation.
- Review the Documents folder on managed devices for a file called svchost.exe, and look for a registered service named Cloudflared.
- Correlate ui_access_control.log with C:ProgramDataGetSupportService_N-CentralLogsBASupSrvc_*.log.gz on Windows endpoints for unauthorized Take Control sessions.
- Investigate sessions tied to apparent N-able support identities such as mspsupport@n-able.com.
- Match network logs against the six IP addresses N-able has published.
Those Take Control logs also appear during legitimate use, so their presence alone is not proof of compromise. Reviewing them helps reduce the risk that an unauthorized session goes unnoticed.
Four questions stay open. How many customers were breached, how many downstream devices the attackers reached, when exploitation began, and who is behind it. N-able has not said whether any data was taken.
SQ Magazine’s Takeaway
The costly detail here is the shape of the persistence. Revoking access at the RMM server assumes the server is the only route in. Cloudflared as a registered service breaks that assumption, because it runs on someone else’s machine, survives a reboot, and reports to an edge network no firewall rule was written to stop. The same trusted-connector pattern showed up when a compromised integration reached Zscaler and other downstream tenants.
What’s next: N-able has promised further updates and has not answered questions about the incident’s scope or the incomplete patch. A CISA catalog entry is plausible given the agency acted on N-central twice before. Huntress framed the hardest call plainly. “Turning off N-central is a significant decision, and it should be made based on risk, not panic,” its analysts wrote. For MSPs that cannot patch fast, the question is whether central visibility is worth the blast radius it creates.
