If your nginx has HTTP/2 switched on and you're running anything from 1.9.5 to 1.31.8, update to 1.30.7 (stable) or 1.31.9 (mainline) and restart the service today. If you can't update right now, turn HTTP/2 off in your server blocks until you can. A reload is not enough, because it does not load the new binary: you need a full restart or a USR2 binary upgrade. This is the fix for NGINX2Root, the set of ten security issues nginx disclosed on 18 November 2026, headlined by two HTTP/2 bugs. Researchers turned the worst of them into unauthenticated remote code execution against stock nginx serving nothing but static files.
What is NGINX2Root?
Nginx is the web server and reverse proxy that sits in front of a huge share of the sites on the internet, often doing TLS termination and load balancing even when something else (Apache, a PHP-FPM pool, a Node app) is doing the actual work behind it. On 18 November 2026 nginx shipped 1.31.9 and 1.30.7 fixing ten security issues at once. Security firm Xint.io, which reported four of the five headline bugs, named the group NGINX2Root, and the name is doing its job, because the worst of these can take an attacker from an unauthenticated HTTP/2 request to code execution on the server.
Wiz Research, Ikotas Labs, Metnir, Rapid7 and Xint each found the core HTTP/2 flaw independently. Wiz has said it used AI models to help validate the bug's impact and frames this, much like PHP2Shell the month before, as evidence that AI tooling makes old memory-safety bugs in widely deployed C code easier to find and weaponise. Expect more disclosures like this, not fewer.
Which nginx versions are affected?
Both primary CVEs sit inside HTTP/2 handling. The higher-priority one, CVE-2026-123007, is an integer underflow (CWE-191) that has lived in nginx's codebase since HTTP/2 support first shipped in version 1.9.5, back in 2015. So that is over a decade of exposure in code most operators had no reason to question.
| CVE | What it is | Affected versions | Severity |
|---|---|---|---|
| CVE-2026-123007 | Integer underflow in HTTP/2 flow control; a cancelled stream refunds bytes to the send window without checking the HTTP/2 maximum, a later bounds check wraps around, and a negative length reaches a memory copy in the TLS send path | 1.9.5 to 1.31.8 | CVSS 4.0 9.2 (Rapid7); nginx.org rates it major |
| CVE-2026-123633 | Use-after-free in HTTP/2, needs proxying to a backend while the worker is running low on worker_connections | 1.19.7 to 1.31.8 | CVSS 4.0 9.2, CVSS 3.1 8.1 per the CVE record; nginx.org rates it medium |
| CVE-2026-121810 | Use-after-free in HTTP/3 | 1.25.0 to 1.31.8 | CVSS 4.0 9.2 per Wiz |
| CVE-2026-123634 | Use-after-free in the resolver | 0.6.18 to 1.31.8 | CVSS 4.0 9.2 per Wiz |
| CVE-2026-113425 | Use-after-free in the mail proxy module | 0.5.33 to 1.31.8 | CVSS 4.0 9.2 per Wiz |
| CVE-2026-118899 | Infinite loop reading a request body over TLS 1.3 | 1.13.0 to 1.31.8 | CVSS 4.0 8.7 per Wiz |
The remaining four fixes cover lower-severity issues in try_files with alias, PROXY protocol v2, stream ssl_preread, and SNI client-certificate validation. Ten fixes landed in one release, but the two HTTP/2 bugs are the ones to prioritise.
Here's where the fixes landed:
- nginx Open Source: 1.31.9 mainline, 1.30.7 stable
- NGINX Plus: 37.3.0.1 and 37.0.8.1 fix CVE-2026-123007; 37.2.0.1 and 37.0.7.1 fix CVE-2026-123633
- NGINX Plus R33 through R36: no fix available, per Rapid7 and Upwind
- NGINX Ingress Controller: fixed in 5.8.0
- NGINX Gateway Fabric: fixed in 2.9.0
Upwind also flags that nginx forks (including OpenResty, Tengine, Angie and freenginx) carry the same vulnerable code and need their own tracking and patching. nginx's own advisory does not cover them.
How dangerous is it really?
Two separate research teams built working exploits against stock configurations. Rapid7 demonstrated unauthenticated RCE against a Docker image of nginx 1.31.8 with HTTP/2 and TLS enabled. Wiz demonstrated RCE with an ASLR bypass against official 1.31.7 and 1.31.8 builds serving only static files (with no backend application in the chain). That is about as close to "nginx itself is the target" as it gets.
Pulling it off is not trivial. Rapid7's writeup says the RCE path needs roughly 128 GB of HTTP/2 traffic sent over a single connection. Over a 10 Gbit/s link that is about two minutes, over 1 Gbit/s about seventeen minutes, and below roughly 300 Mbit/s the transfer takes longer than nginx's default one-hour connection lifetime, so the attack fails. Upwind puts a worker crash at around 60 GB, a lower bar that makes an outage easier to cause than code execution.
The bug sits in the TLS send path, so it only triggers over HTTP/2 served with TLS. If a CDN or layer-7 load balancer in front of your nginx terminates HTTP/2 itself and rebuilds the frames before they reach your origin (Upwind notes that CloudFront, ALB and Cloudflare all work this way), it likely breaks the exploit chain. Treat that as reduced risk, not a fix. Your origin nginx is still vulnerable and still needs patching.
Containers do not remove the issue. Upwind reports researchers chaining CVE-2026-123007 with an unpatched Linux kernel SCTP bug to move from an nginx worker inside a Docker container straight to root on the host. So if you treated the container boundary as your security boundary, this chain shows why it is not one.
As of the cPanel advisory roundup on 20 November 2026, no confirmed attacks in the wild had been reported. That is the current state, not a guarantee. Upwind says Rapid7 published a crash trigger and a full RCE chain on GitHub (lowering the barrier for anyone aiming at unpatched servers).
How do I check whether I'm affected?
First check whether HTTP/2 is enabled. It is off by default (the stock nginx.conf on AlmaLinux 8, 9 and 10 does not enable it), but most HTTPS sites running nginx do turn it on, often long after anyone remembers making that choice. Run this on the server:
sudo grep -RnE '^\s*(listen\s.*\bhttp2\b|http2\s+on)' /etc/nginx
No output means neither HTTP/2 flaw is reachable on that server. If you get a match, patch or disable HTTP/2.
To check a live site from outside, without server access:
curl -so /dev/null --http2 -w '%{http_version}\n' https://www.example.com/
A result of 2 means HTTP/2 is negotiated. A result of 1.1 means it is not (or that something in front of the origin, such as a CDN or load balancer, is terminating it first).
One catch: distribution packages backport security fixes without changing the version string, so nginx -v can still show the old number after you are patched. Compare the package release instead, with rpm -q nginx on RPM-based systems or the Debian/Ubuntu package version with dpkg -l nginx (checked against your distribution's advisory), rather than trusting nginx -v alone.
If your stack does not run nginx at all, such as an Apache-only server or LiteSpeed, none of these CVEs apply to you. Check the front end properly, though. Many shared hosting stacks put nginx in front of Apache as a reverse proxy or static-file cache without making that obvious from the outside, so do not assume you are clear just because your app runs on Apache.
How do I fix it?
Update to 1.30.7 or 1.31.9, or your distribution's equivalent patched package, and restart nginx. Avoid the common mistake of using reload: a reload (nginx -s reload or systemctl reload nginx) does not load the new binary into running worker processes. You need a full restart (or a USR2 binary upgrade if you cannot accept downtime).
If you cannot update immediately, disable HTTP/2 as a stopgap. On nginx 1.25.1 and later, add this inside the affected server blocks:
server {
listen 443 ssl;
http2 off;
...
}
On older versions, remove http2 from the listen line itself instead. Either way, validate the config first:
nginx -t
Then restart, not reload, so no lingering worker process keeps serving HTTP/2 to existing connections. Browsers will fall back to HTTP/1.1 without issue. There is one exception: if you run a gRPC frontend, you cannot disable HTTP/2 this way because grpc_pass depends on it (making patching the practical option there).
Disabling HTTP/2 only closes the two HTTP/2 bugs. It does nothing for the resolver, mail proxy and HTTP/3 fixes in the same release, so treat it as a stopgap while you schedule the actual update, not a substitute for it.
How did hosting platforms respond?
- AlmaLinux, 18 November 2026: backported fixes across every supported nginx stream on AlmaLinux 8, 9, 10 and Kitten 10, pushed straight to production repos ahead of Red Hat's own advisory. Retired streams, including nginx:1.14 which is the AlmaLinux 8 default, got nothing, so check what stream you're actually on (AlmaLinux's NGINX2Root post).
- HestiaCP, 18 November 2026: version 1.10.8 bundles updated nginx 1.30.7 (release notes).
- Ubuntu, 19 November 2026: fixed packages shipped across every supported release from 16.04 through 26.10 under USN-9159-1 (Ubuntu's advisory).
- Plesk: ships nginx twice, once for hosted websites (sw-nginx) and once for the Plesk panel itself (sw-cp-server). The fix landed in Plesk Obsidian 18.0.82 Update 4 and 18.0.81 Update 9. Before that, Plesk's documented mitigation was
plesk bin http2_pref --disablefor websites plus manually disabling HTTP/2 on panel port 8443 (Plesk's advisory). - cPanel: patched via EasyApache, moving ea-nginx to 1.31.9. Anything on ea-nginx 1.31.8-1.2.1 or older is affected, and the interim mitigation was removing the ea-nginx-http2 module (cPanel's November advisory roundup).
If you run Kubernetes, pay close attention: the community kubernetes/ingress-nginx project enables HTTP/2 by default via use-http2: true, bundles nginx 1.27.1, and is an archived project that will not receive a fix. Set use-http2 to false in the controller's ConfigMap now and start planning a migration (because there is no patch coming for this one).
What if I'm on shared or managed hosting?
You cannot patch your host's nginx yourself, but you can still check the exposure. Run the external curl check against your own domain to see whether HTTP/2 is in play. Check your host's status page or changelog for any mention of NGINX2Root, nginx 1.30.7, nginx 1.31.9, or ea-nginx/sw-nginx updates (depending on what control panel they run). Then ask support two direct questions: is your nginx on 1.30.7, 1.31.9, or a distribution package confirmed patched against these CVEs, and when was it actually restarted, not just updated in a repository. A host that cannot give a clear answer within a day is telling you something.
If you are weighing a move to something you control yourself, our VPS hosting rankings are a practical starting point, and our security directory covers WAF and malware-scanning tools worth adding regardless of who manages your server. For the broader picture on what actually stops attacks like this reaching your site, see our guide to what actually protects your hosting. And if you run WordPress specifically, our secure WordPress hosting guide covers the stack-level decisions that matter alongside server patching.
What are the detection signals?
Upwind's writeup lists these signals, none of which show up where most teams look first:
- A single TLS connection carrying tens of gigabytes inbound with very little outbound traffic, the shape of someone building up to the overflow
worker process exited on signal 11in the nginx error log, a segfault consistent with the memory corruption pathworker_connections are not enough, reusing connections, which is the specific precondition CVE-2026-123633 needs to be reachable- An nginx worker process spawning any external program, which it has no legitimate reason to do
The trigger frames themselves never appear in the access log, so access-log-only monitoring will miss this. The evidence sits in the error log and in connection-level telemetry, not request logs.
Frequently asked questions
Is CVE-2026-123007 the same thing as NGINX2Root?
CVE-2026-123007 is the highest-impact bug in the set. NGINX2Root is the name Xint.io gave to five flaws disclosed on 18 November 2026: the two HTTP/2 bugs CVE-2026-123007 and CVE-2026-123633, plus the HTTP/3, resolver and mail proxy use-after-free bugs fixed in the same release.
Does disabling HTTP/2 fully protect me?
It closes CVE-2026-123007 and CVE-2026-123633, the two bugs researchers turned into working RCE demonstrations. It does nothing for the HTTP/3, resolver or mail proxy use-after-free bugs fixed in the same release, so treat disabling HTTP/2 as a temporary measure while you schedule the actual update, not a long-term answer.
My nginx -v still shows the old version after I patched, did it work?
Probably yes. Distribution security fixes are backports, which means your vendor patches the vulnerable code without changing the version string that nginx -v reports. Check the package release number instead, with rpm -q nginx or your distro's package manager, against the vendor's published advisory, rather than relying on the nginx version banner.
Is this being actively exploited right now?
As of the cPanel advisory roundup on 20 November 2026, no confirmed attacks in the wild had been reported. Rapid7 has published a crash trigger and a full RCE chain on GitHub, so the technical barrier to exploitation is low for anyone who wants to try (confirmed exploitation or not). Patch on the assumption that exploitation is coming, not on the assumption that it has not started.
Follow HostList for new rankings, original research, and changes across the hosting industry.



