NGINX2Root brings together five NGINX vulnerabilities, but they do not form a five-step attack. Each concerns a separate route into a worker process. The full root-access demonstration adds a different vulnerability in the Linux kernel, making the operating-system boundary a separate part of the claim.
Xint's original disclosure appeared on 18 November 2026. That day, NGINX released stable 1.30.7 and mainline 1.31.9. Independent lab reports make the story more useful than a list of affected versions: they establish which outcomes have been reproduced and where the evidence stops.
HTTP/2 is only part of the exposure
The upstream security advisory list identifies five different defects within this disclosure:
- CVE-2026-123007: memory corruption in HTTP/2 flow control.
- CVE-2026-123633: a use-after-free in HTTP/2.
- CVE-2026-123634: a use-after-free in the resolver.
- CVE-2026-121810: a use-after-free in HTTP/3.
- CVE-2026-113425: a use-after-free in the mail proxy.
Feature use determines which path exists. Xint describes the second HTTP/2 issue and the HTTP/3 issue in upstream-proxying configurations, the resolver issue with runtime DNS resolution and subrequests, and the mail issue with HTTP-based authentication. Turning off HTTP/2 would therefore address only two of these five paths. It would not settle exposure in a mail service or a proxy using another affected feature.
This also limits what a product inventory can establish. “NGINX installed” is too broad to prove an exploitable path; “HTTP/2 absent” is too narrow to dismiss the entire disclosure. The relevant evidence combines the software build, enabled features and the traffic actually reaching that process.
The underlying software is also different from the Apache HTTP Server HTTP/2 vulnerability disclosed earlier this month. A shared protocol name does not make the fixes interchangeable.
Independent tests demonstrated different outcomes
Censys published its own isolated-lab results. Its compact cleartext HTTP/2 test confirmed the flow-control accounting error for CVE-2026-123007 on NGINX 1.31.8; patched 1.31.9 rejected the sequence. That particular test produced neither a crash nor remote code execution.
For two other paths, Censys obtained worker crashes on unmodified 1.31.8 images: HTTP/2 upstream proxying and mail authentication. Its resolver result needed more qualification. AddressSanitizer confirmed a use-after-free, while crashes came from a custom debug build rather than the stock image. Censys did not reproduce the HTTP/3 issue and did not demonstrate remote code execution for those three use-after-free cases.
Rapid7 separately demonstrated remote code execution through CVE-2026-123007. Its target was an official Linux/amd64 Docker image running NGINX 1.31.8 with HTTP/2 and TLS enabled. The published strategy depends on a particular glibc memory-copy implementation selected using AVX-512 CPU features. Those are conditions of that exploit strategy, not a universal definition of which NGINX systems contain the bug.
These results are complementary. A small protocol-state test can establish that a fix changes behavior without reproducing the full execution chain. A crash establishes an availability impact but does not, by itself, prove that an attacker controls execution. Rapid7's execution result supplies stronger evidence for one flaw on its stated target; it does not convert every other lab crash into a working exploit.
Worker execution and host root are separate claims
Xint reports command execution with the NGINX worker's privileges. Its root demonstration combines CVE-2026-123007 with a separate Linux SCTP kernel vulnerability, including container escape in its test environment. Censys explicitly states that it did not verify that kernel flaw or the combined chain.
The worker account is therefore a consequential boundary. Execution there could expose the data and credentials accessible to that process, or allow malicious use of connections it is permitted to make. Those consequences depend on the deployment's permissions and connectivity. They do not require host-root access to matter.
Host root is a stronger result. It requires the additional escalation path to work against the target kernel and environment. A published Docker demonstration is not evidence that every container runtime, kernel build or managed cloud service can be escaped in the same way. Treating worker compromise and kernel compromise as separate findings makes the reported severity easier to assess without minimizing either.
Fixed releases answer the upstream question
NGINX's 18 November release record states that 1.30.7 and 1.31.9 contain fixes for ten vulnerabilities, including these five. NGINX2Root is a subset of that day's security release, not the name for every issue fixed in it. The upstream advisory entries identify both release lines as fixed for all five.
Distribution packages need their own evidence. Red Hat explains its backporting practice: a fix can be applied to an older packaged version without adopting the latest upstream release number. Consequently, a banner comparison alone cannot reliably establish the patch status of a distribution build. The package maintainer's advisory and CVE coverage are more informative than a bare upstream-looking number.
Nor should these server fixes be confused with the maintenance status of community Ingress NGINX. That controller's retirement is a separate support question. A fix in the NGINX server project does not establish that a particular controller image has incorporated it.
A scan result labelled “NGINX2Root” needs a CVE and a reachable feature behind it. A remediation record needs package evidence covering the applicable flaws. Neither a successful HTTP/2 test nor a fixed version banner answers the separate question of kernel escalation. Those are different findings with different evidence requirements, even when a single headline groups them together.

