F5 released NGINX 1.30.7 and 1.31.9 on 18 November 2026 to fix CVE-2026-123633 and CVE-2026-123007, two NGINX HTTP/2 vulnerabilities that let an unauthenticated attacker run code. Both score 9.2 on CVSS 4.0, and neither has been reported as exploited. If your WordPress site sits behind NGINX with HTTP/2 on, the fix belongs to whoever runs that server.
Key takeaways
- CVE-2026-123633 and CVE-2026-123007 are unauthenticated RCE bugs in NGINX HTTP/2, both CVSS 4.0 9.2, fixed on 18 November 2026 in 1.30.7 stable and 1.31.9 mainline (Plus: 37.0.7.1/37.2.0.1 and 37.0.8.1/37.3.0.1).
- A clean WordPress or Joomla site does not help: Wiz achieved code execution on NGINX serving only static files.
- The same F5 release fixes eight more NGINX CVEs, three of them critical, so take the full update.
- If you cannot upgrade, disable HTTP/2, reload, and confirm old worker processes have exited. No in-the-wild exploitation had been reported at publication.
What are the NGINX HTTP/2 vulnerabilities CVE-2026-123633 and CVE-2026-123007?
Both are memory-safety bugs in NGINX’s own HTTP/2 code, not in any plugin, theme or application behind it. Wiz Research published the details on 18 November 2026, the same day F5, which owns NGINX, released fixes.
- CVE-2026-123007 is an integer underflow in how NGINX tracks HTTP/2 flow control, affecting 1.9.5 through 1.31.8. The flaw dates to 1.9.5, released in 2015, and went unnoticed for more than a decade. Wiz triggered it over HTTPS on 64-bit systems serving nothing but static files.
- CVE-2026-123633 is a use-after-free in HTTP/2 connection handling, present from 1.19.7 through 1.31.8, nearly six years by Wiz’s count. It needs NGINX working as a reverse proxy while it runs short of connection slots.
Both are rated Critical at 9.2 on CVSS 4.0. Neither needs a login. Both need HTTP/2 enabled on a listener the attacker can reach, so switching HTTP/2 off closes both paths at once.
NGINX does not enable HTTP/2 by itself; an administrator or control panel has to turn it on per listener. That is little comfort: most hosting stacks and panel templates enable it for HTTPS sites because it speeds up page loads.
Exposure covers managed WordPress hosts, VPS builds with NGINX in front of PHP-FPM, Docker ingress containers, and shops whose origin answers HTTP/2 directly. Apache-only hosting is out of scope unless NGINX sits in front of Apache as a proxy, a setup many control panels offer.
Five teams share credit: Wiz Research, Ikotas Labs, Metnir, Rapid7 and Xint. F5 told Wiz another team had already reported the use-after-free before Wiz filed on 21 October. When five teams land on the same old bug within weeks, assume people who don’t publish were looking too.
How do these NGINX HTTP/2 bugs work?
HTTP/2 lets a browser send dozens of requests over one connection, so the server keeps detailed books on every stream. Both bugs sit in that bookkeeping. In C, bookkeeping errors become memory corruption.
The flow-control underflow (CVE-2026-123007)
HTTP/2 gives each connection and stream a byte budget, a flow-control window, so neither side floods the other. In Wiz’s account, when a stream is cancelled, NGINX returns its unused allowance to the connection’s window without checking the maximum HTTP/2 allows. A later check subtracts from an unsigned counter, which wraps around, and the bad value drives a memory copy NGINX should never make. No application code takes part, which is why a static file server is enough.
The connection use-after-free (CVE-2026-123633)
NGINX keeps a list of idle connections it can close when it needs room. An HTTP/2 connection joins that list when it has no active requests and should leave when a request starts. Wiz found that a request built from a partially received HEADERS frame never takes its connection off the list.
Under slot pressure, typically while NGINX opens a connection to a backend, cleanup can pick that busy connection and close it. An error path while preparing a GOAWAY frame then frees memory the original request is still using.
Why should a website owner care if the site code is clean?
A fully patched WordPress or Joomla install does not help, because the attacker never talks to WordPress or Joomla. Wiz demonstrated code execution through CVE-2026-123007 on NGINX serving only static files, with no backend application and no third-party module.
NGINX also tends to be shared. Hosts and agencies run one instance as reverse proxy and load balancer for several sites, APIs and admin panels. One vulnerable deployment is one front door for all of them, and the process holding it sees every request, including login forms and session cookies.
Wiz put the impact plainly: “Successful exploitation could let attackers take control of affected NGINX servers, steal sensitive data, tamper with websites, or disrupt services.” For a shop, tampering could mean a skimmer injected into checkout pages at the proxy while every file on the site stays untouched. A scan of the WordPress file tree would come back clean, so check pages from outside, the way a visitor sees them, as well as files on disk.
Memory protections are no shield. Wiz says either bug can be used to read memory before attempting code execution, and its first report to F5 included a confirmed RCE with an ASLR bypass.
One limit: NGINX typically runs a privileged master and less-privileged workers, often as nginx or www-data, so code in a worker is not automatically root. It still gets everything the worker can read and every request it handles, which is plenty for credential theft or page tampering.
Which NGINX versions did F5 fix on 18 November 2026?
NGINX Open Source is fixed in 1.30.7 stable and 1.31.9 mainline. Anything from 1.9.5 through 1.31.8 is in range for at least one bug. NGINX Plus has separate fixed builds per CVE, with R33 and 37.0.0 as the first affected versions. F5’s NGINX security advisories for the 18 November release list each CVE.
| Product | CVE | First affected | Fixed in |
|---|---|---|---|
| NGINX Open Source | CVE-2026-123633 | 1.19.7 | 1.30.7 stable, 1.31.9 mainline |
| NGINX Open Source | CVE-2026-123007 | 1.9.5 | 1.30.7 stable, 1.31.9 mainline |
| NGINX Plus | CVE-2026-123633 | R33; 37.0.0 | 37.0.7.1, 37.2.0.1 |
| NGINX Plus | CVE-2026-123007 | R33; 37.0.0 | 37.0.8.1, 37.3.0.1 |
Take the whole release. It fixes eight more NGINX CVEs, three of them also critical at 9.2:
- CVE-2026-121810, a use-after-free in HTTP/3 on proxy configurations (1.25.0 through 1.31.8).
- CVE-2026-123634, a use-after-free in the DNS resolver that allows code execution when an attacker controls an upstream host name (0.6.18 through 1.31.8).
- CVE-2026-113425, a use-after-free in the mail proxy module, exploitable when NGINX proxies mail with auth_http (0.5.33 through 1.31.8).
The other five score lower: CVE-2026-118899 (possible denial of service on servers supporting TLS 1.3, 8.7), CVE-2026-120993 (denial of service with proxy_protocol v2), CVE-2026-125283 (a buffer overflow in an uncommon alias plus try_files setup), CVE-2026-123640 (SSRF-like access via ssl_preread) and CVE-2026-120967 (a client certificate bypass in specific SNI setups).
The disclosure moved fast: Wiz reported the use-after-free on 21 October, F5 sent a patch for review on 10 November, Wiz reported the flow-control bug on 16 November, and F5 folded it into the 18 November release.
One caveat on version numbers: Linux distributions often backport fixes while keeping an older upstream number, so nginx -v showing 1.24.0 on Debian or Ubuntu may already be patched. Debian, Ubuntu, Red Hat and others ship those packages on their own schedules, and we cannot tell you when each will land. Check your distribution’s security tracker by CVE number, and disable HTTP/2 until the package arrives.
How do you check HTTP/2 and write the host ticket?
On shared or managed hosting, the NGINX in front of your site is almost never yours. As with our Apache HTTP/2 host check, your job is to ask precise questions and get evidence back.
Quick checks you can run yourself
- curl –http2 -sS -o /dev/null -w ‘%{http_version}’ https://yoursite.example/ prints 2 if your site negotiates HTTP/2. Browser developer tools show the same in the Protocol column.
- Look at the Server response header. Many hosts hide it, so a missing header proves nothing.
- This only describes the first hop. If a CDN answers, its HTTP/2 says nothing about the NGINX at your origin.
If you have shell access
- Dump the effective configuration with nginx -T and search it for http2. Look for both the older listen … http2 form and the http2 on directive used from 1.25.1 onward. Servers configured across several NGINX generations often contain both.
- Record nginx -v and nginx -V, then check the package with dpkg-query -W ‘nginx*’ or rpm -qa ‘nginx*’ against your distribution’s advisory.
- HTTP/2 is not limited to TLS. NGINX can serve cleartext HTTP/2 (h2c), so a TLS-only test can miss exposure.
What to put in the host ticket
- Is the edge, the origin, or both running NGINX, and is HTTP/2 enabled on them?
- Is the NGINX build below 1.30.7 stable or 1.31.9 mainline (or the Plus builds listed above), or does your distribution package include the fixes for CVE-2026-123633 and CVE-2026-123007?
- Please upgrade, reload, and confirm that no worker processes from the old binary are still running.
- Please send the nginx -v output after the change.
Ask for that last string explicitly. A control panel version or stack image tag is not the NGINX version, and “we’re up to date” is not evidence.
Where audits usually go wrong
When we audit NGINX setups, the misses rarely happen at the version string. They happen around it.
- Containers left behind. Patching NGINX on a Docker host does nothing for the NGINX inside your containers; the image has to be rebuilt or pulled and the container redeployed. We regularly find a patched host with an outdated ingress container doing the real work.
- Mixed syntax. Someone edits listen lines, and an http2 on directive in an included file keeps HTTP/2 alive. A single listen 443 ssl http2 in one vhost turns it back on for every server block sharing that address and port. Read nginx -T output, not nginx.conf.
- Forgotten vhosts. Staging sites, admin panels and old origin hostnames often run on the same NGINX with nobody watching.
- CDN bypass. A site behind a CDN often still answers on its origin IP. Test that origin directly or restrict it to the CDN’s address ranges.
- Evidence lost to cleanup. If you see worker process exited on signal 11, keep the error logs and any core dumps before restarting anything. Wiz calls unexpected worker crashes and commands launched by NGINX general warning signs, not confirmed indicators of these CVEs, but they justify a closer look. So do shells or outbound connections owned by the NGINX worker user.
Version and config checks show potential exposure, not compromise. Do not run proof-of-concept memory corruption code against production. If you suspect the front-end server was abused, scan the sites behind it for injected scripts with SiteGuarding Antivirus and rotate credentials that passed through the proxy.
How do you disable NGINX HTTP/2 if the host cannot upgrade today?
Turn HTTP/2 off on every listener until the fixed build is running. Wiz’s advice is direct: “If upgrading is not immediately possible, temporarily disable HTTP/2.” Both CVEs require it, so this closes both.
- On NGINX 1.25.1 and newer, set http2 off or remove http2 on from each server block.
- On older configs, remove the http2 flag from every listen line, including those in sites-enabled, conf.d and control panel snippets.
- Test the config with nginx -t, then reload.
- Confirm with curl that the site now answers on HTTP/1.1.
Test first: some APIs and gRPC services depend on HTTP/2 and will break without it, and a busy site may load slightly slower over HTTP/1.1. That trade is fine for a few days.
The reload step trips people up. A graceful reload starts new workers but lets old ones finish their connections, and until they exit they run the old configuration. Long-lived connections can keep them alive for a while. Compare worker PIDs and start times with ps before and after; a clean systemctl reload message is not proof.
The workaround does nothing for the other CVEs in the release, such as the HTTP/3 and DNS resolver bugs. On your own VPS, install the fixed package from your distribution or nginx.org, run nginx -t, reload, confirm no old workers remain, check nginx -v, then re-enable HTTP/2 and repeat the worker check.
A website firewall or CDN reduces exposure if it terminates HTTP/2, talks HTTP/1.1 to your origin, and the origin cannot be reached directly. It does not replace the upgrade, because origins are often reachable by IP address.
Are these NGINX flaws exploited, and what is still unknown?
No in-the-wild exploitation had been reported at publication, and Wiz’s disclosure mentions no attacks. That may not last: Wiz published a root-cause analysis on patch day, five teams found the bugs independently, and plenty of NGINX servers update slowly. Plan for working exploits within days, not months.
Wiz credits AI-assisted scanning. It ran Wiz Atlas, the tool behind its AI Code scanning product, with Gemini 4 Carbon and ChatGPT 5.6 Sol for validation. Wiz argues that several teams finding the same long-dormant flaws in PHP and NGINX shows discovery is speeding up: “Complexity can no longer be relied upon as a barrier to discovery or exploitation.” That cost is falling for attackers too.
We covered a similar race earlier this month with Apache’s HTTP/2 flaw CVE-2026-119618. Web server bugs now go from disclosure to public detail in days, and a quarterly patch cycle is too slow for that. Ask your host how many days it takes them to ship a critical web server fix.
Still open: whether working exploits circulate privately, how quickly distributions and managed hosts ship backported packages, and which control panels bundle their own NGINX builds. F5’s NGINX security advisories remain the authoritative fixed-version list.
FAQ
What is CVE-2026-123633 in NGINX?
A use-after-free in NGINX's HTTP/2 connection handling, affecting 1.19.7 through 1.31.8. It can lead to unauthenticated code execution when NGINX acts as a reverse proxy under connection-slot pressure.
Does my WordPress site need a patch, or does my host?
The host, or whoever runs the NGINX server. Updating WordPress, plugins or themes does not fix bugs in NGINX itself.
How do I check if NGINX has HTTP/2 enabled?
From outside, curl –http2 against your site prints 2 if HTTP/2 is negotiated. On the server, search nginx -T output for both listen … http2 and http2 on.
Are these NGINX flaws being exploited in the wild?
No in-the-wild exploitation had been reported at publication, but Wiz's detailed write-up and five independent discoveries mean exploits could follow quickly.
If you are not sure what sits in front of your site, or you want a second set of eyes after your host patches, our website protection package adds a firewall layer and monitoring that flags injected code on your pages.
