twimco.Let’s talk ↗︎
Menu
Field notes / Product & engineering

NGINX fixed two bugs today that let attackers in without a login. Patch before Black Friday, then prove it's running

Somewhere in front of your site, NGINX is probably answering every request, and it just got two critical HTTP/2 fixes. The update is the easy part. Making sure the new code is really running, everywhere NGINX hides, is where teams slip.

Two engineers at a studio workbench at night pull a rack server halfway out to restart it, with a laptop, a printed checklist, a brass padlock, a wrench and a desk clock nearby, and lit holiday storefronts outside the window.

Patch now, freeze or not. NGINX 1.30.7 and 1.31.9, released November 18, 2026, fix ten bugs, including two HTTP/2 flaws (CVE-2026-123007, CVE-2026-123633) that can let an unauthenticated attacker run code. No attacks are reported yet, but exploit code for one is public. Find every copy, including inside ingress controllers and gateways. Update and restart, not reload, or turn HTTP/2 off until a fix ships. Then confirm the new binary is running.

Your website's front door got a security fix today

There's a good chance your website runs on NGINX even if nobody on your team has ever said the word. It's the program that answers first when a browser asks for your site. It handles the HTTPS, then hands the request to your store, app or API. A lot of hosting setups, Kubernetes clusters and API gateways put it there by default.

On Wednesday, November 18, F5 released NGINX 1.30.7 (stable) and 1.31.9 (mainline), with fixes for ten security bugs. Two of them are the reason for this note. Both are in how NGINX handles HTTP/2, the newer version of the web protocol most browsers use. CVE-2026-123007 goes back to version 1.9.5 from 2015. CVE-2026-123633 shows up when NGINX passes traffic on to an app behind it. F5 scores both 9.2 out of 10 (CVSS 4.0). Wiz and the other teams that found them say an attacker who can reach the server can run code on it without logging in. Wiz showed it working on a server that was only serving ordinary static files.

As of this evening, nobody has reported attacks. CSO Online cites CISA's assessment that neither bug has been exploited in the wild, and neither is on CISA's list of actively exploited vulnerabilities. But Sophos says proof-of-concept exploit code for the first one is already public. And most of these bugs were found by researchers working with AI tools: Xint, Wiz, Metnir and Ikotas Labs all used them, and Rapid7 found the same flaw on its own. If five teams found the same bugs on their own, assume someone less friendly can too.

Black Friday is November 27, and your code freeze has probably started. Like the WordPress security release we wrote about earlier this evening, this belongs on the short list of things a freeze should let through. The update itself is the easy part. What worries us is a team that applies it, ticks the box and stays exposed anyway.

Three ways to be "patched" and still exposed

The version number says one thing, the package says another. Linux distributions often fix a bug in their own build without changing the NGINX version number, so nginx -v can't tell you on its own. Xint points out that Ubuntu 22.04 still reports 1.18.0, even though its package had reached 1.18.0-6ubuntu14.22 through security updates by late October. The real answer is in the full package version and your distribution's advisory. Tonight there's a twist. When we checked, around 7:40 PM Eastern, the Debian and Ubuntu security trackers listed NGINX as vulnerable on every supported release. If your servers use the distribution's package, a routine apt upgrade doesn't fix this yet.

The new code is on disk, but the old code is still running. A reload (nginx -s reload, or "reload" in a hosting panel) re-reads the configuration. It doesn't swap out the program that's already running. Sophos says to restart NGINX after updating so the patched program loads. Xint says the same, or do a proper live binary upgrade. Miggo Security's CEO told CSO the real risk is "how many teams will think they've patched when they haven't."

It's NGINX, but nobody calls it that. OpenResty, Kong, Apache APISIX, NGINX Gateway Fabric and the community Kubernetes ingress-nginx controller all build on NGINX code. According to CSO, Metnir reproduced the second bug in OpenResty and crashed the shipped versions of Kong, APISIX, NGINX Gateway Fabric and ingress-nginx, all of which turn HTTP/2 on by default. Each one gets fixed only when its own maintainer ships an update. And the community ingress-nginx project was archived on March 24, 2026. Its README says there won't be any more security fixes. Same story for NGINX Plus R33 through R36, which F5 no longer supports: no fix is coming.

Decision guide

Looks done vs. actually done

ConsiderationLooks doneActually done
Versionnginx -v shows a number you think is fine.Full package version matches your distro's advisory, or the build is 1.30.7 / 1.31.9 or later.
UpdatePackage installed, then reloaded.Restarted, and the running process isn't pointing at a deleted binary.
ScopeThe web server everyone knows about.Every copy: containers, ingress controllers, gateways, appliances, hosts' and vendors' setups.
Stopgaphttp2 off in one server block.Every block on that address and port, checked from outside with -alpn h2.
Decision framework based on the Xint, Sophos and CSO guidance for the November 18, 2026 NGINX release. Illustrative, not client data.
A reload re-reads your settings. It doesn't swap out the program that's already running.

Find every NGINX, including the ones inside other things

Start with a list, not a terminal. Write down everything that answers web traffic for you: servers, containers, Kubernetes clusters, the API gateway, the admin panel on that appliance in the closet, and anything a hosting company or CDN runs for you. Then work through it.

  • Servers you run. nginx -v gives the version. Then look up the full package version (dpkg -l nginx or rpm -q nginx) and compare it with your distribution's advisory.
  • Containers. docker run --rm --entrypoint nginx <image> -v prints the NGINX inside an image. An image built in March still has March's NGINX, whatever the host has installed since.
  • Kubernetes. kubectl exec <controller-pod> -- nginx -v on your ingress controller pods. If the answer is community ingress-nginx, this is the week to plan the move off it.
  • Things somebody else runs. Managed hosting, a load balancer, a CDN or a platform you pay for. Ask them directly: "Do you run NGINX in front of our site, and when will the 1.30.7 or 1.31.9 fixes be live?"

Then sort by risk. The two headline bugs need HTTP/2. Xint published three quick checks: nginx -V shows whether the HTTP/2 module is built in, nginx -T shows whether any server turns it on, and an openssl s_client ... -alpn h2 test from outside your network shows whether the internet can actually reach it. If a CDN sits in front, run that last test against the origin server too. A CDN that handles HTTP/2 itself may soften the attack, Wiz says, but that depends on the setup. An origin server people can reach directly is still exposed.

Anything that's internet-facing, speaks HTTP/2 and runs an old build goes to the top. That's tonight's work.

Patch tonight, or switch HTTP/2 off tonight

For each box at the top of the list, you're choosing between two moves.

Patch, if a fixed build is available to you. NGINX's own package repositories already carry 1.30.7 and 1.31.9, and the official Docker images for 1.30.7 were published this afternoon. NGINX Plus customers on the current lines move to 37.3.0.1 or 37.0.8.1. If you already install from nginx.org or the official images, this is a normal update, and it brings in every fix from this release at once.

Switch HTTP/2 off, if it isn't. That's the case if you depend on your distribution's package and it hasn't shipped yet, if a vendor hasn't released a fix, or if you can't restart until a quieter hour. On NGINX 1.25.1 or later, set http2 off; in each affected server block and remove http2 from the listen lines. On older versions, removing http2 from listen is enough. Then test the site, reload, and run the outside check again. Wiz, Xint and Sophos all describe this as a stopgap. It blocks the two HTTP/2 bugs. Turning off HTTP/3 (the quic listeners) blocks a third. It does nothing for the rest, and anything that needs HTTP/2, like native gRPC, will stop working. For ingress-nginx, Rescana's suggestion is use-http2: "false" in the controller ConfigMap.

Don't swap a distribution package for nginx.org's packages in the middle of a freeze just to get the fix tonight. That changes file paths, modules and service scripts at the same time. Switch HTTP/2 off tonight, and patch properly once your distribution ships, or once there's time to test the move on a staging copy you trust.

Decision guide

Patch tonight, or switch HTTP/2 off tonight?

ConsiderationPatch tonight whenSwitch HTTP/2 off tonight when
Where NGINX comes fromYou already install from nginx.org or the official Docker images.You rely on Debian or Ubuntu packages, which had no fix this evening.
Embedded copiesThe vendor has shipped a fixed release.The vendor hasn't, or the project is retired, like community ingress-nginx.
Restart windowYou can restart in a quiet hour tonight.You can reload safely but can't restart until later.
What depends on HTTP/2gRPC or other services need it.Browsers and clients will fall back to HTTP/1.1 without trouble.
Decision framework: assess these conditions for each box you run. Illustrative, not client data.

Then prove the new code is running

This is the step people skip, and it's why "we updated" and "we're fixed" can drift apart.

  1. Restart in your quietest hour. systemctl restart nginx, or rolling your pods or containers onto the new image. Xint warns that a restart drops open connections, so expect a short blip. Tell whoever's watching orders and support.
  2. Check the program that's running, not the file on disk. Find the master process with pgrep -f 'nginx: master', then run ls -l /proc/<pid>/exe. On Linux, if that link ends in "(deleted)", the process is still running a binary that has since been replaced on disk. In other words, the old code. Comparing the master's start time (ps -o lstart= -p <pid>) with when the package was installed tells you the same thing.
  3. In Kubernetes, check what's actually running. kubectl rollout status should finish, and nginx -v inside a fresh pod should show the fixed build.
  4. If you switched HTTP/2 off, test from outside. The -alpn h2 check should no longer come back with h2, and the site should still load normally over HTTPS.
  5. Watch the error log for a few days. Look for lines like worker process ... exited on signal. CERT-EU and Xint both say a crash isn't proof of an attack, but repeated crashes on a box that was exposed are worth a closer look.
  6. Write it down. One line per box in the freeze log: what changed, when, who checked, and whether HTTP/2 is still off and needs turning back on.

This is the same "how you know you're done" step from our Java4Shell note yesterday. CSO counts this release among a run of AI-discovered holes in PHP, Node, OpenSSH and Java, including PHP2Shell last month. Expect more weeks like this one.

If you'd rather someone else did this tonight

None of this is exotic. It just lands on a busy week, it touches machines people would rather not think about, and the step that matters most, proving the fix is live, is easy to forget once the update says "done."

If your site or app runs behind NGINX, or you're not sure what sits in front of it, that's the kind of production work our web application team does: keeping the systems around your software current and checked, not only the code. Tell us what you're running and we'll help you find every copy, patch or switch it off, and get it written down before Black Friday.

Related work: Sutton|Place.

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Sutton Place ↗︎