Nginx HTTP/2 RCE (CVE-2026-123633 & CVE-2026-123007): Affected-Servers Checklist and Caddy/Traefik Migration Guide
Two critical NGINX HTTP/2 remote code execution flaws disclosed on November 18, 2026 require every self-hoster to check their servers this week. On November 18, 2026, Wiz Research โ together with Ikotas Labs, Metnir, Rapid7, and Xint, who independently uncovered the same flaws โ published details of two critical remote code execution vulnerabilities in NGINX, disclosed in coordination with F5. Both are unauthenticated, both are rated Critical with a CVSS 4.0 score of 9.2, and both live in NGINX's HTTP/2 handling: your sites can be exposed even when your application code is secure, and exploitation does not require a backend application at all โ the researchers demonstrated code execution on NGINX serving ordinary static files. NGINX is the world's most widely used open-source web server, fronting more than 225 million websites. If you self-host anything โ a content site, an API, a reverse proxy in front of your app stack โ this is a this-week problem, not a this-quarter problem. This guide walks through how to determine whether your servers are affected, what to do in the next hour if you can't patch immediately, and how to think about the two migrations most teams are quietly evaluating right now: mitigating in place, or replacing NGINX with Caddy or Traefik.Quick Verdict
| Question | Answer |
|---|---|
| Am I affected? | If your NGINX serves traffic with HTTP/2 enabled on an unpatched build, yes โ check version and config today. |
| How bad is it? | Critical (CVSS 9.2), unauthenticated RCE. An attacker who succeeds can run code as the NGINX process user, steal data, tamper with sites, or pivot inward. |
| What's the fix? | Upgrade to the November 18, 2026 F5 security release (or your distro's backported package). |
| Can't patch today? | Disable HTTP/2 on affected server blocks as a temporary mitigation. |
| Should I switch to Caddy or Traefik? | Maybe โ but patch first. A migration is a project; the vulnerability is live now. |
What Was Disclosed
Two separate vulnerabilities, both requiring HTTP/2 to be enabled on the affected server: CVE-2026-123633 โ a use-after-free in NGINX's HTTP/2 implementation. Wiz reported it to F5 on October 21, 2026, with confirmed remote code execution including an ASLR bypass and a proposed fix. F5 confirmed another research team had independently reported the same flaw. CVE-2026-123007 โ an integer underflow in HTTP/2 stream handling that has been present in the code since NGINX 1.9.5, released in 2015 โ meaning it sat in one of the internet's most scrutinized codebases for over a decade. Wiz's testing demonstrated code execution against a stock NGINX serving static files, with no backend and no third-party modules involved. A few structural points worth absorbing:- The flaws are in NGINX itself. Your application framework, your code quality, your WAF rules upstream โ none of them change the exposure. The front door is the problem.
- HTTP/2 enabled is the trigger condition. Servers with HTTP/2 off (or fronted by a CDN that terminates HTTP/2 upstream of an HTTP/1.1-only NGINX) are not reachable through these flaws.
- One NGINX often fronts many apps. A single compromised edge node can put every service behind it at risk, which is why teams are treating this as infrastructure-level, not app-level.
Step 1: Are You Affected? (15 Minutes)
Work through this checklist on every server you operate or are responsible for: 1. Inventory your NGINX instances. Don't assume you know where all of them are โ marketing sites, internal tools, staging boxes, and that one VPS from two years ago all count. A quickps aux | grep nginx or package query (dpkg -l | grep nginx, rpm -qa | grep nginx) on each host is a start.2. Check the version.
nginx -v gives the version; nginx -V shows the build. Compare against your vendor's November 18, 2026 security advisory. If you run NGINX through a Linux distribution package, your distro may have backported the fixes with its own version string โ check the distro security tracker rather than the upstream version number alone.3. Confirm whether HTTP/2 is actually enabled. Version alone doesn't tell you exposure; configuration does. Look for
http2 on any listen directive, or http2 on; directives in your server blocks:
listen 443 ssl http2;
http2 on;
If neither appears anywhere in your configuration, HTTP/2 is off and these vulnerabilities don't apply to that server โ though "HTTP/2 off in config" and "we think it's off" are different statements; verify on the actual files.4. Check what's in front of your NGINX. If a CDN or load balancer terminates HTTP/2 and speaks HTTP/1.1 to your origin NGINX, the origin isn't exposed via these flaws. If TLS with HTTP/2 passes through to NGINX (common with pass-through load balancing), it is.
5. Managed and container images. If you run the official NGINX container images, pull the November 18, 2026-or-later rebuilds and redeploy. Same for any base images your team bakes โ a cached
nginx:latest from three months ago is a vulnerable one.
Step 2: Mitigate Today If You Can't Patch
If an immediate upgrade isn't possible โ change freeze, vendor appliance, embedded device โ reduce exposure now:- Disable HTTP/2 on affected server blocks. Remove the
http2parameter fromlistendirectives or sethttp2 off;, then reload NGINX. Clients will fall back to HTTP/1.1. You lose multiplexing on that server until you re-enable it after patching, which is a real but acceptable cost against unauthenticated RCE. - Restrict what the edge can reach. Ensure the NGINX process user has no write access it doesn't need, no access to secrets beyond what the served sites require, and โ where feasible โ no route to internal networks the public sites don't need. Containerized or sandboxed edge deployments pay off here.
- Watch for exploitation noise. Your WAF, CDN, or edge logs may show probing against HTTP/2 endpoints in coming weeks. This class of flaw attracts scans quickly after disclosure.
- Confirm upstream vendor status. If NGINX is bundled inside an appliance, managed hosting plan, or PaaS you don't fully control, open a ticket and ask explicitly which release fixes CVE-2026-123633 and CVE-2026-123007 and when it lands.
Step 3: Should You Replace NGINX With Caddy or Traefik?
This question comes up after every major NGINX event, and the honest answer is: patch first, evaluate second. But the evaluation is worth doing on the merits, because the two candidates genuinely solve different problems. Caddy is the strongest "replace the edge server" candidate for most small teams. Its signature feature โ automatic HTTPS via ACME, on by default, renewed without you thinking about it โ removes an entire category of edge misconfiguration. Configuration is short and declarative, which shrinks the surface for the kind of subtle directive interactions that linger in decade-old NGINX configs. It's written in Go with a memory-safe runtime, and HTTP/2 and HTTP/3 are handled by its underlying library. The migration friction is real but modest: re-express your server blocks as Caddyfile blocks, validate redirects and header rules one by one, and keep NGINX in front (or behind) during a testing window. Traefik is the better fit when your "edge" is really a dynamic container platform. If your services come and go under Docker, Kubernetes, or a Nomad-style scheduler, Traefik's discovery-based routing โ labels and providers instead of hand-maintained upstream blocks โ matches how that infrastructure actually behaves. For a traditional static site plus a few apps on fixed hosts, it's more machinery than you need. A few honest caveats before you migrate:- Different bugs, different era. Moving off NGINX does not buy you a bug-free edge; it buys you a different codebase with a different (and shorter) vulnerability history, plus operational traits that happen to reduce certain risk classes. Presenting Caddy or Traefik as "secure where NGINX is not" overstates the case.
- Your muscle memory doesn't migrate. Whoever owns the new config owns its incidents. If your team has a decade of NGINX operational knowledge and zero days of Caddy, you've traded a known quantity for an unfamiliar one. Budget the learning curve into the decision.
- Do it as a project, not a fire drill. A panicked weekend migration is how header rules, redirect chains, and cache behaviors silently break in production. Use this event as the trigger to start a proper evaluation, with a test host and a cutover plan.
The Longer-Term Lesson
Two of the most-deployed pieces of internet infrastructure โ NGINX here, PHP with PHP2Shell last month โ have shipped decade-old memory-safety flaws that multiple research teams found within the same window, with AI-assisted analysis lowering the cost of finding them. Three practical takeaways for a small team: 1. Shrink your edge's privileges. Whatever terminates traffic should be able to do as little as possible: read-only where feasible, no secrets it doesn't strictly need, no lateral network access beyond its role.2. Make patching boring and fast. The teams that patched within 24 hours weren't braver; they had inventory, staging mirrors, and a tested upgrade path. If "upgrade the edge server" is a scary sentence in your organization, that's the gap to close.
3. Keep an exit option warm. Knowing how you'd move to Caddy, Traefik, or a managed CDN-fronted architecture โ even if you never do โ is cheap. Rehearsing it once is cheaper than discovering it under pressure.
Sources and Verification
- Wiz Research: Unauthenticated Remote Code Execution in NGINX (CVE-2026-123633 and CVE-2026-123007) โ primary disclosure, November 18, 2026
- F5/NGINX security release of November 18, 2026 โ apply the latest release for your packaging channel and verify against your distribution's security tracker
- Coverage context: Hacker News front-page discussion, November 18, 2026