
Two flaws in NGINX’s HTTP/2 module, CVE-2026-123007 and CVE-2026-123633, can allow remote code execution (RCE) in an NGINX worker before the client authenticates. The first releases that fix both are 1.30.7 stable and 1.31.9 mainline for NGINX Open Source, and 37.0.8.1 LTS and 37.3.0.1 CR for NGINX Plus. Wiz Research’s November 18, 2026 disclosure · NGINX Plus release notes
CVE-2026-123007 can affect static file servers as well as reverse proxies. HTTP/2 flow control limits how much data a sender can transmit before the receiver is ready for more. NGINX can miscalculate that allowance and corrupt a worker’s memory while sending a response. Wiz demonstrated RCE in its lab on a 64-bit NGINX server serving ordinary static files over HTTP/2 and TLS. Include download mirrors and documentation sites in your inventory. The affected Open Source code dates back to 1.9.5. F5’s flow-control advisory · Wiz’s static-file test
CVE-2026-123633 affects reverse proxies that accept HTTP/2 from clients. When a worker has few connection slots left and forwards a request to a backend, it can free memory that the request still needs. This use-after-free can crash the worker or allow code to run with the worker’s permissions. The affected Open Source code dates back to 1.19.7. F5’s reverse-proxy advisory
Use the release listed for your branch or a later patched release on that branch. NGINX Plus operators on 37.0.7.1 or 37.2.0.1 already have the reverse-proxy fix and still need the flow-control update. NGINX Open Source security advisories · NGINX Plus release notes
| Release line | First release with both fixes |
|---|---|
| NGINX Open Source stable | 1.30.7 |
| NGINX Open Source mainline | 1.31.9 |
| NGINX Plus Long-Term Support (LTS) | PLS.37.0.8.1 |
| NGINX Plus Continuous Release (CR) | PLS.37.3.0.1 |
If an ingress controller or appliance supplies NGINX, look up the fixed controller image or appliance release from that vendor. For a Linux distribution package, use the vendor’s full package revision to account for backported fixes. How security backports work
Run nginx -T on the host or inside the container, using the same configuration path as the service. It prints the configuration and its included files. Look for http2 on; or the older http2 parameter on a listen directive. For each listener that accepts HTTP/2, record which clients can reach it and which package or container image serves it. NGINX command-line reference · NGINX HTTP/2 module
Deploy the patched binary. Restart NGINX on the fixed package, or roll out a patched container image. Confirm that every serving replica runs the expected build and that the old workers have exited. For an autoscaled service, verify that the deployment template starts new replicas with the patched image. NGINX process and binary upgrade guide
If the upgrade has to wait, temporarily disable HTTP/2. Check which clients require it first, especially native gRPC clients. On NGINX 1.25.1 and later, set http2 off; for each affected server. On all versions, remove the http2 parameter from those servers’ listen directives. Validate with nginx -t, reload, and wait for the previous workers to exit. Using a client that offers both protocols over TLS, verify the listener selects HTTP/1.1. Restore HTTP/2 after verifying the patched rollout. NGINX HTTP/2 directives · NGINX gRPC module
Consider a public download mirror. Visitors need to fetch files, a release job needs to publish them, and an on-call engineer needs to patch the hosts. The NGINX update fixes the public listener. Adaptive can give the release job and the engineer separate, narrowly scoped access.
Keep the release console private. If the team has an internal console for approving releases, configure its private origin to accept connections only from Adaptive’s broker. Release managers open the console from Adaptive’s SSO catalog and use it through an isolated browser. Their laptops need only a connection to Adaptive. Adaptive isolated browser sessions
Keep publishing credentials off the mirror. Give the release job its own workload identity in Adaptive, scoped to the host where it publishes files. Adaptive brokers the job’s connection and manages the downstream credential in its vault. Mount the release directory read-only for NGINX. That protects the stored files even if a compromised worker alters a response on its way to a visitor. Adaptive workload identities
Give the on-call engineer a patch window. For example, approve 30 minutes of SSH access to the affected mirror hosts. The engineer signs in with SSO and MFA, updates NGINX, and verifies the running build. Adaptive records the session and expires the grant. If a worker exits during maintenance, the incident reviewer can match the restart command in the recording to the timestamp in the NGINX error log. Just-in-time infrastructure access · Session audit and playback
Q: Does a CDN or WAF protect an NGINX origin from these flaws?
A CDN or web application firewall can reduce exposure if it terminates HTTP/2 and opens an HTTP/1.1 connection to the origin. The browser’s h2 indicator describes its connection to the edge. Check whether another route, such as an internal load balancer, can still reach an HTTP/2 listener on NGINX. Wiz’s analysis of CDNs and WAFs
Q: How do I check an Ubuntu NGINX package for both CVEs?
Look up your Ubuntu release in each CVE record. When Ubuntu lists a fix, compare its full package revision with the revision installed on the serving host or in the container. The two CVEs have separate status records. Ubuntu’s CVE-2026-123007 tracker · Ubuntu’s CVE-2026-123633 tracker
Q: Does a configuration reload activate the patched NGINX binary?
A configuration reload starts workers from the master process already in memory. To load a new executable, restart the service on the patched package or follow NGINX’s live binary upgrade procedure. During a live upgrade, verify that workers from the old executable have exited before closing the change. NGINX binary upgrade procedure
Q: Does an NGINX worker crash prove that someone exploited these vulnerabilities?
A worker can crash for several reasons. Wiz recommends investigating unexplained crashes and commands launched by NGINX. Compare the NGINX error log, host or container process events, and deployment timestamps to establish what happened on that instance. Wiz’s mitigation and detection guidance

