What happened
On November 18, 2026, F5 issued its out-of-band notification, NGINX published fixed Open Source releases, and CERT-EU advised rapid updating of internet-facing systems. The five code-execution paths are CVE-2026-123007 with HTTP/2 enabled, CVE-2026-123633 with HTTP/2 and an upstream server, CVE-2026-121810 with HTTP/3 and an upstream server, CVE-2026-123634 with resolver and request-derived proxying conditions, and CVE-2026-113425 in a mail proxy using proxy on with auth_http.
NGINX Open Source 1.31.9 and 1.30.7 are fixed; CERT-EU lists NGINX Plus 37.0.0 through 37.0.7.1, 37.1.0 through 37.2.0.1 and R34 through R36 as affected. The Open Source ranges vary by flaw: CVE-2026-123007 affects 1.9.5 through 1.31.8, while the other paths begin at different historical releases and extend through 1.31.8.
Xint’s exposure workflow uses nginx -v 2>&1, nginx -V 2>&1 | grep -q – ‘–with-http_v2_module’, sudo nginx -T 2>&1 | sed ‘s/#.*//’ | grep -nE ‘listen[[:space:]].*http2|^[[:space:]]http2[[:space:]]+on[[:space:]];’, and an external openssl s_client test with -alpn h2. CERT-EU recommends checking NGINX error logs for worker process … exited on signal and correlating crashes with process-creation evidence because a crash alone is not proof of code execution.
No in-the-wild exploit IP addresses, domains or file hashes were published by the cited authorities at publication time. Attribution posture: No actor attribution has been established because CERT-EU reported no in-the-wild exploitation at publication time.
Why this matters now
The disclosure affects the data plane of a widely embedded web server, reverse proxy, ingress layer and load balancer. The most exposed path, CVE-2026-123007, requires only HTTP/2 on a listening port. Other paths depend on less common configurations, but they are independent; disabling HTTP/2 does not close the resolver, HTTP/3 or mail-proxy issues.
Version banners are not decision-grade evidence because Linux distributions and appliances may backport fixes without changing the upstream NGINX version. Conversely, a recent upstream version can remain affected. Owners need the full package revision, vendor advisory, loaded binary and running configuration before marking an asset safe.
Researchers demonstrated code execution and a separate host-escalation chain in controlled conditions, while CERT-EU reported no exploitation in the wild. That combination supports emergency preventive work and focused hunting without claiming compromise. Internet-facing systems deserve priority, followed by ingress controllers, appliances and products that bundle NGINX.
The decision for security leaders
Assign the exposure decision to infrastructure and application-platform owners jointly. Network scans can identify listeners, but only configuration and package evidence can determine whether HTTP/2, HTTP/3, resolver, proxy or mail-module conditions are present and whether a distribution backport is installed.
Prioritise the actual internet path. Test both public front doors and the origin NGINX listener behind CDNs, managed load balancers or reverse proxies. A successful HTTP/2 negotiation at the CDN does not prove the origin is exposed, while a protected public hostname does not prove direct origin access is blocked.
Require process replacement as part of remediation. Installing a package or reloading configuration may leave old worker binaries or established connections in service. The closure record should show the running binary, completed restart or supported live upgrade, protocol state and vendor-specific package evidence.
Evidence of closure
- The asset register records each NGINX package revision, module set, listener and owning service.
- Running Open Source binaries report fixed or vendor-backported builds after restart or supported live upgrade.
- External protocol tests confirm the intended HTTP/2 and HTTP/3 exposure state at origin listeners.
- Hunt results document the disposition of abnormal worker exits and NGINX child-process events.
The Security.io assessment
This is a high-impact preventive vulnerability decision rather than a confirmed enterprise incident. CERT-EU states that researchers demonstrated CVE-2026-123007 but that no exploitation in the wild had been reported. ExploitationStatus is therefore proof-of-concept, not confirmed-active.
The five flaws should be handled as a configuration family, not a single HTTP/2 bug. Disabling HTTP/2 can reduce exposure to CVE-2026-123007 and CVE-2026-123633, but it does not address the HTTP/3, resolver or mail-proxy paths. Organisations that close the ticket after one protocol change may retain a separate vulnerable condition.
No in-the-wild exploit IP addresses, domains or file hashes were published by the cited authorities at publication time. That limits retrospective hunting to behavioural evidence such as abnormal worker exits, child processes and unusual connection patterns. A clean hunt strengthens closure but cannot replace verified patch or mitigation state.
Questions for the morning meeting
- Can infrastructure teams identify the full package revision rather than relying on the NGINX banner?
- Which listeners expose HTTP/2 or HTTP/3 directly rather than through a managed proxy?
- Which deployments use request-derived DNS resolution, mirroring or the NGINX mail proxy?
- Can owners restart or live-upgrade processes without unacceptable service interruption?