21
N-central servers Huntback found running a miner on 12 Nov 2026
19
of those 21 already on the Hotfix 2 build
33
alleged N-central flaws dumped on GitHub with no vendor notice
7
CVEs Hotfix 2 fixed; 4EverAble2Shell is not one of them
An RMM server is every client at once
A remote monitoring and management server is not one more box on a managed service provider's network. It is the box that runs scripts, installs software and opens remote sessions on every machine the provider looks after, across every client it serves. Whoever holds root on it does not need a foothold in each client. They already have one, installed and trusted.
That is the lens for this week. On 12 Nov 2026 a public proof of concept called 4EverAble2Shell gave unauthenticated attackers root on N-able N-central servers that have the Legacy API switched on. Within hours, Huntback reported a Monero miner running on 21 live N-central servers, and 19 of them were on the Hotfix 2 build N-able had shipped two days earlier as its critical security update. There is no fix for this chain yet.
So the build number did not decide who was hit. The configuration did. A feature N-able ships off by default and does not recommend was on, and the server answered requests from the internet. Those two conditions are older than this bug and will outlast it. A miner is the cheapest thing a root shell on an RMM server can do. The same access reaches every client endpoint the server manages, which is why we treat the RMM as part of each client's attack surface, not only the provider's.
What happened to N-able N-central this week
N-central is a remote monitoring and management platform used by managed service providers and internal IT teams. It runs as a hosted service from N-able and as an on-premises appliance that providers operate themselves.
The week started with a dump. Late on Sunday 8 Nov, US time, an independent researcher published more than 30 alleged zero-day vulnerabilities in N-central to a GitHub repository without notifying the vendor, according to BleepingComputer. Tenable counts 33 alleged flaws in the repository's README, grouped into four exploit chains, and dates the earliest commit to 04:14 UTC on 9 Nov. BleepingComputer reports the researcher said the work was AI assisted. N-able called it "non-responsible communication of critical vulnerabilities" and took Hosted N-central offline on 9 Nov as a precaution, while saying it had not observed active exploitation.
On 10 Nov N-able released N-central 2026.4 Hotfix 2, build 2026.4.1.26. Its release notes call it "a critical security update". Tenable lists the 7 CVEs it fixes, rated 8.3 to 9.3 on CVSS 4.0: an XML external entity flaw and a server side request forgery at 9.3 each, an OGNL injection at 8.7, two local privilege escalations, a file upload path traversal that leads to code execution, and a stored cross site scripting flaw. The highest are CVE-2026-125883 and CVE-2026-125881. The release also temporarily disables N-central's built-in remote control. Tenable notes N-able has not mapped those CVEs to the 33 findings in the dump, so nobody outside N-able can say which of the researcher's claims Hotfix 2 closed.
On 12 Nov the same researcher published two more sets: 4EverAble2Shell against the N-central server and Agent2Shell, which The Hacker News says chains more than 10 zero-day vulnerabilities against Windows agents. That is the chain this post is about.
-
Product //
N-able N-central, on-premises servers -
Result //
Unauthenticated command execution as root -
Condition //
Legacy API enabled, which is off by default -
Hit builds //
2026.4.1.26 (Hotfix 2) on 19 of 21, per Huntback -
Fix //
None yet, per Tenable -
Identifier //
No CVE assigned -
Exploited //
Yes, per Huntback; N-able says it has seen no compromise
The claims about exploitation do not all agree, and the disagreement matters. N-able, quoted by The Hacker News, says hosted environments are not affected by this issue and that it has seen no signs of compromise. Tenable reports that Huntress had not observed exploitation in its partners' environments, and that Previdian and Beazley Security Labs saw attempts against decoys. Huntback, publishing on 12 Nov, reported something stronger: a miner running on live customer servers. Neither Tenable nor The Hacker News has independently confirmed Huntback's count. We report it as Huntback's finding, because it is specific, and because the indicators it published let any operator check their own server.
How 4EverAble2Shell reaches root
The chain is short, and we describe it only at the level of what each stage gives the attacker. It starts at the Legacy API, an older SOAP interface the server still ships.
Figure 01
From one request to every client
- 01
In //
A request to the Legacy API, no login.
- 02
Run //
Commands as the N-central service account.
- 03
Root //
A privileged helper lifts it to root.
- 04
Stay //
A miner under systemd or cron.
- 05
Reach //
The console that manages every agent.
Result //
One server, and the trust every client gives it.
Stage one needs nothing but network reach to an N-central server with the Legacy API enabled. The proof of concept's own notes say the API must be on, and N-able told customers it is "turned off by default in the product and is not recommended". Previdian traces the code execution to unsafe class handling during SOAP fault processing in the bundled Apache Axis component. Stage three, in Previdian's account, is a separate weakness: privileged helper functions on the appliance that turn a service level foothold into root.
Stage four is what Huntback found on the 21 servers: XMRig, the common Monero miner, using about half the CPU cores, kept alive with a systemd unit named xvd.service or with cron entries, and reporting to its pool under worker names that start with nc_ followed by the victim's address. Huntback says the miner is the only payload it has seen. Previdian reports the payload also tried to fetch and run another shell script.
Stage five is the one nobody has reported, and the one we care about. An N-central server holds the configuration, credentials and agent channels for every device it manages. Root on that server is root on the thing that tells those agents what to run. A miner wastes that position. A ransomware crew would not. We have written about this shape before: Xen Orchestra this week was a console that owns every virtual machine beneath it, and the Qubole AWS roles were third party trust that outlived the relationship. An RMM server is both at once.
Why Hotfix 2 did not end it
Three facts keep this from being a patch and move on story.
The first is the build. Huntback writes that "19 of the 21 report build 2026.4.1.26" and puts it plainly: "Applying Hotfix 2 did not protect them." Hotfix 2 fixed seven named CVEs. 4EverAble2Shell was published two days after it and is not among them. Teams that patched on 10 Nov and closed the ticket were exposed on 12 Nov, and a version scanner would have marked them green.
The second is the speed. Huntback saw the first check against a server from a single address at 08:30 UTC on 12 Nov, and the full chain from 09:16 UTC. By 12:50 UTC it counted 21 mining servers: 9 in the United States, 4 in South Africa, 4 in Canada, and one each in the United Kingdom, Germany, France and Sweden. The sources disagree on exactly when the proof of concept went up that morning, either 03:58 or 07:13 UTC, but on either reading the gap between publication and compromise was hours. There was no patch window to miss. There was only the configuration already in place.
The third is that there is still no fix. Tenable says N-able has not released one for the 12 Nov claims. N-able's own update says it is "restricting certain N-central features and integrations" while it develops further remediation. Per Tenable, that includes switching off the Intune integration in hosted environments and deactivating it for on-premises customers who use it, and N-able recommends keeping an on-premises server offline until it is upgraded. When a vendor turns features off across its customer base, that is a statement about how much of the product it trusts this week.
There is also history. Tenable lists five earlier N-central CVEs in the CISA Known Exploited Vulnerabilities catalog, three of them added in 2026: CVE-2026-18577 on 3 Aug, CVE-2026-18556 on 4 Aug and CVE-2026-86218 on 8 Sep. None of the seven Hotfix 2 CVEs was in the catalog as of 12 Nov, and 4EverAble2Shell has no CVE to list. A program that keys on CVE IDs or on KEV has nothing to match against this campaign. We made the same point about the OpenSSH bypass on Alpine: the feeds lag the exploitation, and an RMM server will not wait for a number.
How to detect N-central compromise
Start on the server. Huntback's indicators for this campaign are concrete: an unexpected systemd unit named xvd.service, a miner process, cron entries the team did not write, files under /tmp/.xvd or /var/tmp/.xvd, and outbound connections to 157.20.241[.]16 and to mining pools. If your N-central appliance ships host telemetry anywhere, the rule below is a starting point.
Figure 02
N-central miner artifacts from the 12 Nov 2026 campaign
Sigma
title: N-central server miner artifacts, 4EverAble2Shell campaign
status: experimental
description: Files Huntback lists for the XMRig miner installed on N-central servers on 12 Nov 2026
references:
- https://huntback.io/blog/ncentral-4everable2shell-miners-on-live-servers
logsource:
product: linux
category: file_event
detection:
selection_dir:
TargetFilename|contains:
- '/tmp/.xvd'
- '/var/tmp/.xvd'
selection_unit:
TargetFilename|endswith: '/xvd.service'
condition: selection_dir or selection_unit
falsepositives:
- Unlikely on an N-central appliance
level: high Then read the web logs in front of the server. BleepingComputer and Huntback both point to POST requests to /dms/services/ServerUI, the Legacy API path, as the thing to investigate. If the Legacy API is meant to be off and that path has been answering, that alone is a finding. Look back to 8 Nov at least, because the earlier chains were public from then.
Then look at the network. A management server that can open connections to mining pools or to arbitrary addresses on the internet has egress it does not need. Block the address Huntback lists and alert on any outbound connection from the appliance that is not to N-able's update service or your own agents.
The miner is the loud part. Do not let it set the scope. Huntback's advice is the right one: "Treat a hit as a full compromise, not a cleanup job." A root shell on the server could have read the credentials it stores and pushed jobs to agents before or after the miner went in. A hit means reviewing scheduled jobs, script runs and remote sessions pushed to client devices since 8 Nov, rotating every credential the server holds, and rebuilding the appliance rather than cleaning it.
How we test an RMM platform on an engagement
We do not need this chain to ask the question it asks. On an external test, the first thing we settle is whether the RMM console answers on the internet at all, and which of its interfaces answer. An admin page or an old API that responds without a login is something an external test identifies from its responses, without logging in; our exposed admin and debug endpoint checks show how, and atarus-recon on our open source page covers the wider external inventory.
Inside, with credentials the client issues for the test, we map what the RMM can do. Which technicians can push a script to every device. Which accounts are shared. Whether a single technician login reaches every client or only the ones that technician serves. Whether the RMM's own credentials can also delete the backups. We report the number of steps from the RMM console to a domain controller at a client. Where a technician can push a script to every device, the answer is one.
For organisations whose fleet is run by a provider, we treat the provider's RMM as in scope for the conversation even when it is out of scope for testing. The questions are short. Which RMM runs our devices, and is it hosted or on premises? Is the Legacy API, or its equivalent, switched on? Can the console be reached from the internet? What can a technician run on our machines without a second person approving it? Would we see it if they did? A provider that can answer those in a day is a provider that has thought about this. One that cannot is the supply chain risk, whatever their certificate says.
Then we check what the client could see. Does anything on the client side log a script that arrives through the RMM agent? Does an alert fire when the agent runs something new at three in the morning? Many teams exclude the RMM agent from endpoint tooling so it can do its job. If yours does, an attacker on the server inherits that exclusion.
What is still coming
Expect a fix and plan for the gap. N-able says further remediation is in development. Until it ships, the controls are the Legacy API switch, network reach and the features N-able has already turned off. Those cost some convenience. Pay it.
Expect more chains. Of the 33 alleged flaws, N-able has publicly tied fixes to seven CVEs, and the researcher has kept publishing. Agent2Shell, aimed at the Windows agents rather than the server, matters to every client endpoint and is not covered by anything in this post. Each new set has come with working code, and the gap from code to exploitation this week was hours.
Expect better operators than this one. A miner on a management server is a low value use of a high value position, and it announced itself through CPU load. The next crew to use the same access may sit quietly and push something to the agents instead. That is the outcome to plan against.
Expect identifiers to follow, and do not wait for them. A CVE and a KEV listing may come for 4EverAble2Shell. Vulnerability scanners will flag builds after that. None of it changes what to do today.
The class will outlive this week. RMM platforms, backup consoles and hypervisor managers keep old interfaces alive for compatibility, and the old interface rarely gets the same review as the new one. The defence is the same every time: legacy features off, the console on a management network, technician rights narrow, and logs held somewhere the console cannot reach.
What to do today
-
Turn the Legacy API off on every on-premises N-central server. It is off by default; confirm it rather than assume it, and record who turned it on and why.
-
Take the N-central console off the internet. Put it behind a VPN or management network, and follow N-able's advice to keep an un-upgraded server offline.
-
Stay on Hotfix 2 and watch for the next fix. Hotfix 2 closes seven CVEs; it does not close 4EverAble2Shell, so a green version scan is not the end of the job.
-
Run the host checks and treat any hit as a full compromise. The xvd.service unit, files under /tmp/.xvd or /var/tmp/.xvd, miner processes and egress to mining pools. On a hit, review jobs pushed to client devices since 8 Nov, rotate every credential the server holds and rebuild.
-
Block egress from the appliance. Allow only what it needs, and alert on anything else; the Sigma rule above is a starting point for the host side.
-
If an MSP runs your devices, ask them today. Which RMM, hosted or on premises, Legacy API state, internet reach, and what a technician can run on your machines without approval.
SOURCES
- N-able N-central 2026.4 HF2 release notes // build 2026.4.1.26, a critical security update, built-in remote control temporarily disabled
- Huntback // XMRig miners on 21 live N-central servers, 19 of them on build 2026.4.1.26, with host and network indicators
- Tenable // the 33 alleged flaws, the seven CVEs and scores fixed in Hotfix 2, no fix yet for the 12 Nov chain, five earlier N-central CVEs in KEV
- BleepingComputer // 4EverAble2Shell is unauthenticated root that needs the Legacy API, and N-able is restricting features while it builds further fixes
- BleepingComputer // N-able took Hosted N-central offline on 9 Nov after more than 30 alleged zero-days were published