November 16, 2026
Nineteen of Twenty-One Were Patched. The Exploit Needed a Switch.
A default is not a boundary. Itβs a starting position that survives exactly as long as nobody has a reason to change it.

By Raj Namdev
7 min read
30-second version:_ On November 12, a mining pool listed 21 rigs named after the N-central servers they were running on. Nineteen of those servers were on build 2026.4.1.26 β Hotfix 2, released two days earlier to fix seven CVEs. The chain that owned them needed a feature the hotfix didn't remove: the Legacy API. N-able's own mitigation for the same API family a year earlier was to leave it off by default. One year later, at least nineteen production servers had it on._
At 12:50 UTC on November 12, the wallet behind an XMRig campaign had 21 rigs submitting shares, each labelled nc_ plus an IP address. Every one of those addresses served N-central's login page. Nine were in the US, four in South Africa, four in Canada, one each in the UK, Germany, France and Sweden.
The same wallet had 631 rigs in total, with prefixes for Jpom, FreePBX, MLflow, Apache Flink, PaperCut, SFTPGo and AJ-Report. N-central was the newest entry on that list and the only one with a management plane behind it.
What the patch fixed, and what the chain used
Hotfix 2 shipped on November 10 with seven CVEs, CVSSv4 from 8.3 to 9.3: an XXE class, a schema-hint SSRF, an OGNL injection, a syslog-ng privilege escalation, a path-traversal RCE, a stored XSS and an insecure IPC channel. That is the list a scanner checks.
The chain that landed on the miners is not on that list. It was published on November 12 at 03:58 UTC as 4EverAble2Shell, and the researcher's own README states the precondition in the second line:
Legacy API needs to be enabled; Forgot this was turned on during earlier testing for other vulnerabilities.
The mechanism, from the same README: the public Axis endpoint constructs a shipped JLine terminal-settings class from attacker-controlled SOAP fault data before validating its type. JLine passes that value into sh -c as the nable user, and a helper chain (setLDFileOwner, then serverControl) makes the last step run as root. One malformed fault body, two shipped classes, uid 0.
Huntback saw the first probe at 08:30 and the full chain at 09:16. That is five hours from public code to weaponised request, and about two and a half hours after that to 21 confirmed hosts.
Nov 8, late β Researcher publishes 33 N-central findings with PoCs Nov 9 β N-able takes hosted N-central offline Nov 10 β Hotfix 2 (2026.4.1.26) ships, seven CVEs assigned Nov 12, 03:58β4EverAble2Shell published Nov 12, 08:30 β First probe against huntback decoys Nov 12, 09:16 β Full chain: Axis fault β nable β root Nov 12, 12:50β21 rigs live in the pool; 19 report Hotfix 2
N-able's position is that the second wave depends on a non-default configuration, with the Legacy API enabled and reachable from the internet. That is accurate and it is also the whole problem: the setting is not default, and nineteen patched production servers still had it.
Counted, not quoted
The "33 vulnerabilities" figure gets repeated without anyone opening the repository, so I did. I cloned MSPNightmare/CVEs at commit main, counted the files, and grepped every service path out of them.
git clone --depth 1 https://github.com/MSPNightmare/CVEs
cd "CVEs/0-Days/N-Central N-Able/Able2Shell"
ls *.py | wc -l # 33
cat *.py | wc -l # 4257
grep -h -oE '/dms2?/services2?/[A-Za-z0-9_]+' *.py \
| sort | uniq -c | sort -rngit clone --depth 1 https://github.com/MSPNightmare/CVEs
cd "CVEs/0-Days/N-Central N-Able/Able2Shell"
ls *.py | wc -l # 33
cat *.py | wc -l # 4257
grep -h -oE '/dms2?/services2?/[A-Za-z0-9_]+' *.py \
| sort | uniq -c | sort -rn/dms2/services2/ServerMMS2 β 9 PoCs β appliance tasking, logging, cross-appliance writes /dms/services/ServerUI β 9 PoCs β service import, sysconfig, asset jobs /dms2/services2/ServerMMS3 β 5 PoCs β patch metadata, ecosystem credentials /dms2/services2/ServerII2 β 5 PoCs β automation policy, device execute /dms2/services2/ServerEI2 β 5 PoCs β registration tokens, technician accounts
Twenty of the 33 scripts reference a /dms path; seventeen name a full endpoint. The other thirteen build their URL from a base argument or hit a non-SOAP route, which is why the counts don't sum to 33. Ten of the 33 filenames start with anonymous_, and in the companion Agent2Shell folder twelve of fourteen do.
That concentration is the finding. Thirty-three findings, four chains, and the paths collapse into five legacy SOAP services that share an authentication model. A patch that fixes sinks inside those services leaves the services.
The credential-free chain is the one worth reading closely, because it shows why. An anonymous MMS2 caller stages XML and JSP content. The service-import parser resolves an external DTD before it checks the UI session, so the attacker's DTD is fetched while the request is still "unauthenticated." That fetch leaks the rotating NCI session, the attacker's JSP is written, and the JSP executes as uid=992(nable).
Order of operations. The entity resolution happens first, the session check happens second. Every XXE fix in the world is downstream of that.
The same endpoint, one year earlier
In August 2025, Horizon3 found sessionHello on /dms/services/ServerMMS. N-central ships with built-in appliances whose values are static across installs, so the method returned a valid appliance session ID to anyone who asked. That was CVE-2025-9316.
Appliance sessions were scoped, so that ID didn't work on ServerUI. But importServiceTemplateFromFile never validated the session at all, and its call chain reached XMLValidator.validateXML(), which built a SAXParser without the secure defaults that disable external entities. That was CVE-2025-11700.
The chain needed one more step to be useful, and it's the step people forget. applianceLogSubmit on the same appliance session wrote attacker-controlled content to /opt/nable/webapps/ROOT/applianceLog/network_check_log_<appliance_ID>.log. The importer takes a file path. So: write your XML to a known location, then ask the importer to parse it, and let the parser resolve the entity against /etc/passwd.
Same endpoint. Same parsing weakness. Different year.
How did N-able close it? In 2025.4.0.9, by making the abused legacy SOAP APIs unavailable in the default configuration. Not removed from the product β turned off, with a switch to turn them back on.
That switch is the thing nineteen servers had flipped. Upgrades preserve configuration. A server migrated from a 2025 build, or one where an integration needed the old API, arrives at 2026.4.1.26 with the feature enabled and the newer CVEs patched.
"Off by default" is a migration state
This is the part worth stealing for your own work, whether you test software or defend it.
A default is not a boundary. It's a starting position that survives exactly as long as nobody has a reason to change it. Legacy APIs exist because something in the customer's stack still calls them, and the customer who turned one on in 2024 does not revisit that decision in 2026. The vendor ships a fix, the scanner goes green, and the reachable surface is unchanged.
N-able's remediation plan makes the same point by reversing the 2025 approach. Hotfix 3 will "permanently remove access to the Legacy API." Delete, not default off. That is a different class of fix, and it is the only one that closes a switch nobody is auditing.
For hunters, the transferable method is one line: when you find a bug on a legacy endpoint, enumerate every method the endpoint exposes and test which ones skip the session check. That is precisely how the 2025 finding worked β most ServerUI methods rejected the appliance session, importServiceTemplateFromFile didn't. One method out of dozens, and it was the one that reached the parser.
Two habits follow from that:
- Get the WSDL or the service listing, not just the endpoints you saw in a request. The unused methods are the ones nobody has tested since the feature shipped.
- Test the order of checks, not just their presence. If the parser runs before the session lookup, an "authenticated" endpoint is unauthenticated for the duration of one request.
The check, and the limits of this piece
If you run N-central, the audit is config-first, because the CVEs are the easy part:
# is the legacy surface reachable at all?
curl -s -o /dev/null -w '%{http_code}\n' \
https://<host>/dms/services/ServerUI
# what the exploit looked like in the logs
grep -i 'exceptionName' /var/opt/n-central/logs/dmsservice_soap.log
# persistence from the observed campaign
ls -la /tmp/.xvd /var/tmp/.xvd 2>/dev/null
systemctl status xvd.service 2>/dev/null
crontab -l 2>/dev/null | grep -i xvkeep# is the legacy surface reachable at all?
curl -s -o /dev/null -w '%{http_code}\n' \
https://<host>/dms/services/ServerUI
# what the exploit looked like in the logs
grep -i 'exceptionName' /var/opt/n-central/logs/dmsservice_soap.log
# persistence from the observed campaign
ls -la /tmp/.xvd /var/tmp/.xvd 2>/dev/null
systemctl status xvd.service 2>/dev/null
crontab -l 2>/dev/null | grep -i xvkeepA 200 on the first line is the finding. Everything else is containment.
What I did not do: run N-central, test the chain, or verify the configuration on any of the 21 hosts. My contribution here is the file census and the 2025-to-2026 endpoint match β both reproducible from public repositories. Huntback did not document whether the Legacy API was enabled on each victim, and N-able disputes parts of the disclosure. The inference that all nineteen had it on rests on the researcher's own statement that the chain requires it and on N-able's confirmation that exploitation depends on that setting.
The other thing I can't tell you is whether those nineteen were compromised before or after the hotfix. Applying 2026.4.1.26 doesn't remove a root shell that was already installed, and the observed payload leaves a watchdog that relaunches the miner every ten minutes if it dies. Patch status is not integrity status.
Twenty-one servers, one setting, and a hotfix that closed seven things next to it.
Sources
All counts and file numbers are from my own run on November 16, 2026 against MSPNightmare/CVEs (main), 0-Days/N-Central N-Able/. No N-central instance was contacted, tested, or attacked.
- Able2Shell README and 33 PoCs β
0-Days/N-Central N-Able/Able2Shell/; 33.pyfiles, 4,257 lines, 10anonymous_names; chain descriptions quoted from the README.github.com/MSPNightmare/CVEs - 4EverAble2Shell README and PoC β the "Legacy API needs to be enabled" line, the Axis/JLine mechanism, and the
setLDFileOwnerβserverControlescalation.github.com/MSPNightmare/CVEs/tree/main/0-Days/N-Central%20N-Able/4EverAble2Shell - Huntback, "4EverAble2Shell is already mining Monero on live N-central servers" β 21 rigs at 12:50 UTC, 19 on build 2026.4.1.26, geography, the 631-rig wallet, the
jup.shinstaller, and the decoy timeline (08:30 probe, 09:16 full chain).huntback.io/blog/ncentral-4everable2shell-miners-on-live-servers - Horizon3, "N-able N-central: From N-days to 0-days" (November 17, 2025) β
sessionHello, CVE-2025-9316,importServiceTemplateFromFile,XMLValidator.validateXML()and the insecure SAXParser, and the 2025.4.0.9 mitigation that made the abused legacy SOAP APIs unavailable by default.horizon3.ai/attack-research/attack-blogs/n-able-n-central-from-n-days-to-0-days/ - N-able, "N-able Vulnerability Remediation Update" (November 13, 2026) β Legacy API "defaulted to off and not a recommended configuration"; Hotfix 3 "will also permanently remove access to the Legacy API."
n-able.com/blog/n-able-vulnerability-remediation-update - N-able, Hotfix 2 CVE list β seven CVEs, CVSSv4 8.3β9.3, released November 10, 2026, build 2026.4.1.26.
- Tenable, "Able2Shell FAQ" β 33 findings, four chains, two reaching root, and the note that huntback did not document the Legacy API setting per victim.
tenable.com/blog/able2shell-frequently-asked-questions-for-reported-n-able-n-central-zero-day-vulnerabilities - BleepingComputer, "New N-able 4EverAble2Shell zero-day exploited hours after PoC released" β 21 systems, 19 patched, the five-hour gap, and N-able's statement about the non-default configuration.