The update is installed. The status page is green. Your browser still says `h2`. On a patch day, those three facts can all be true while an old NGINX process is still serving requests.

On November 18, NGINX released **1.30.7 stable** and **1.31.9 mainline** to fix two HTTP/2 vulnerabilities, CVE-2026-123007 and CVE-2026-123633. If you run NGINX, start by finding the instances that receive HTTP/2, then get the fix from whoever supplies your build and verify that the patched processes are handling traffic. [NGINX security advisories](https://nginx.org/en/security_advisories.html)

That sounds like one job until you ask the person on call who owns the connection. The app team owns the release. The platform team owns the ingress. A provider might own the public edge. Each team can finish its part and leave the question open.

## The vulnerable code can run before your app sees a request

Wiz and other research teams reported the vulnerabilities to F5 before the coordinated public disclosure. Wiz demonstrated remote code execution in a lab against NGINX serving static files for CVE-2026-123007. A backend application was unnecessary in that test. Both vulnerabilities require HTTP/2 on the affected NGINX instance. CVE-2026-123633 concerns deployments that also proxy requests to an upstream server. Check the individual advisories for the full conditions. [Wiz disclosure](https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities) · [F5 advisory for CVE-2026-123007](https://my.f5.com/manage/s/article/K000163926) · [F5 advisory for CVE-2026-123633](https://my.f5.com/manage/s/article/K000163928)

That puts the proxy itself in the incident's scope. A team that has kept its application dependencies current can still have work to do, and several services can share the same affected proxy. Start the inventory with the processes that accept connections. Fill in the applications they serve once you know where the patch has to land.

## Put an owner on each connection

Imagine a browser reaching an API through a CDN and an NGINX ingress. The browser's `h2` tells you what it negotiated with the public edge. The edge may establish a separate connection to the origin and choose its own protocol. Cloudflare, for example, documents a separate HTTP/2-to-origin setting. Check the configuration and ordinary traffic telemetry for each connection your team operates, including any internal ingress or origin listeners that your architecture exposes. [Cloudflare HTTP/2 to Origin](https://developers.cloudflare.com/speed/optimization/protocol/http2-to-origin/)

Write down who receives HTTP/2 and who can change the deployed build. “The platform team” is a useful start. “Priya owns the ingress rollout and will attach the running image digest” is a handoff an on-call engineer can follow. For a managed service, record the provider's advisory or support case and the person on your side who will follow it through.

If an affected listener has to stay on an old build for now, Wiz recommends temporarily disabling HTTP/2 there. Check the service impact first, especially where native gRPC depends on HTTP/2. Give the change an owner and a review time, and verify the intended protocol change on that connection. The other fixes in the same NGINX release still need their own applicability check. [Wiz mitigation guidance](https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities) · [NGINX release notes](https://nginx.org/en/CHANGES)

## Match the fix to the build you actually run

For upstream NGINX Open Source, the advisory lists affected versions starting at **1.9.5** for CVE-2026-123007 and **1.19.7** for CVE-2026-123633. The November 18 upstream releases fix both. NGINX Plus uses separate release tracks. The November 18 builds **37.0.8.1 LTS** and **37.3.0.1 CR** cover both issues. The CVE-2026-123633 fix had already shipped in the October 14 releases for those tracks. [NGINX security advisories](https://nginx.org/en/security_advisories.html) · [NGINX Plus release notes](https://docs.nginx.com/nginx/releases/)

A Linux distribution can backport the fix and keep an older upstream version in its package name. A controller image can bundle its own NGINX build. Use the advisory for the distribution release, controller image, or managed product you deploy. Carry the full package revision or immutable image digest into the rollout so the next person can identify the exact build. [Red Hat on backporting](https://access.redhat.com/security/updates/backporting) · [How to verify a vendor's security backport](/blog/linux-chrome-cve-fix-status)

## Close the ticket when the old workers are gone

A package install changes files on disk. NGINX distinguishes a configuration reload from an executable upgrade, and old workers can keep serving existing clients while they drain. Follow the supported upgrade or replacement procedure for your deployment, then check every serving replica and confirm the old worker generation has retired. Send a normal request through the path customers use. If that path carries gRPC, make an ordinary RPC part of the check. [NGINX process lifecycle](https://nginx.org/en/docs/control.html)

The handoff can fit in a ticket comment. For a hypothetical three-replica API, it might read **“Platform owns the edge-to-ingress HTTP/2 connection. All three serving replicas report the approved image digest. Old replicas have drained. The API and gRPC smoke checks passed through the public endpoint. The CDN provider advisory is tracked in a separate open item.”**

When the next advisory lands, that comment is the start of the investigation. It tells the next person on call which connection to look at, who can change it, and what evidence will let them call the rollout done.