A critical vulnerability has been found in NGINX, the software that underpins websites and APIs around the world, that could allow an attacker to execute arbitrary code on a server without authentication. The issues are two flaws in HTTP/2 handling, CVE-2026-123007 and CVE-2026-123633, collectively known as "nginx2root." On November 18, 2026, F5, the developer of NGINX, released fixed versions that address these and eight other vulnerabilities, for a total of 10.
What makes the flaws especially serious is that they lie not in a web application but in NGINX itself, the software that handles the traffic. In its technical analysis, Wiz Research reports achieving remote code execution even on a setup that only serves static files, with no backend application or external modules. The impact is not limited to dynamic sites such as WordPress; it also extends to static sites that serve HTML and images.
Responses have begun in Japan as well. On November 19, Xserver temporarily disabled HTTP/2 on six of its services and switched them to HTTP/1.1. The measure is a stopgap to reduce the chance of attack until patched versions are applied, and it shows that the vulnerabilities are affecting the operation of large-scale hosting services.
HTTP/2 memory corruption puts even static sites at risk
Both flaws relate to memory management when NGINX processes HTTP/2 traffic. HTTP/2 is a protocol that handles multiple requests and responses in parallel over a single connection. This improves communication efficiency, but it also makes the handling of connection state and data transfer volumes more complex.
CVE-2026-123007: exploitable even when serving static files
CVE-2026-123007 is a memory corruption vulnerability in HTTP/2 flow control. Flow control adjusts how much data is sent according to the receiver's capacity; in HTTP/2, it manages the amount of data that can be sent for the connection as a whole and for each individual stream.
According to Wiz Research's analysis, when queued data is canceled, NGINX returns the send allowance that had been allocated to that data. However, this process does not adequately verify that the result stays within the limit set by HTTP/2, so certain operations can push an internal value into an invalid state. Subsequent processing then compounds this with an integer underflow and conversions between signed and unsigned integers, ultimately causing a memory copy that uses an invalid length.
As a result, the NGINX worker process may crash, and an attacker may be able to read or overwrite data in memory. If an attacker can obtain memory layout information and bypass address space layout randomization (ASLR), arbitrary code execution may be possible.
Importantly, the attack does not require a vulnerability in a web application. Wiz Research demonstrated remote code execution with an ASLR bypass against a setup serving static files over TLS in a 64-bit environment, using official NGINX 1.31.7 and 1.31.8. Because the affected code path is used inside NGINX for ordinary static file delivery, servers are affected even without a backend application or third-party modules.
According to the official NGINX security advisories, the vulnerability has existed since version 1.9.5, released in 2015. That means an externally exploitable memory corruption problem remained in code that has been in use for more than a decade.
CVE-2026-123633: a flaw in connection handling for reverse proxies
The other flaw, CVE-2026-123633, is a use-after-free vulnerability in HTTP/2 connection management. It is a problem when NGINX operates as a reverse proxy, forwarding the requests it receives to a backend application server.
When NGINX runs short of available connection slots, it closes connections judged to be reusable to free up space. Under certain conditions, however, an HTTP/2 connection that is still being processed is treated as reusable, and memory it needs can be freed. If request processing continues after that, NGINX refers to memory that has already been freed, leading to memory corruption and code execution.
This vulnerability is affected when HTTP/2 is enabled and certain other conditions coincide, such as reverse-proxy forwarding and a shortage of connection slots. Unlike CVE-2026-123007, simple static file delivery alone does not create the same attack path. Canonical's vulnerability description likewise lists HTTP/2 client connections and proxying to a backend as conditions for impact.
CVSS 9.2 severity: servers can be compromised without authentication
Both vulnerabilities are rated 9.2 (Critical) under CVSS 4.0 and 8.1 (High) under CVSS 3.1. NGINX's own classification rates CVE-2026-123007 as "major" and CVE-2026-123633 as "medium," but the fact that they can lead to code execution from outside without authentication makes them a serious problem.
If an attack succeeds, code may be executed with the privileges of the NGINX worker process. This could lead to theft of information on the server, defacement of websites, or service outages. Because NGINX is often used as a common foundation for multiple websites and APIs, the compromise of a single server could also harm the multiple services behind it.
However, despite the nickname "nginx2root," exploiting the vulnerabilities does not by itself grant root privileges. The privileges of the executed code depend on those of the NGINX worker process. Moreover, going from memory corruption to actual code execution requires steps such as learning the memory layout and bypassing protection mechanisms, so the difficulty of an attack varies by environment.
Wiz Research has succeeded in attacks in several test environments but has said it will not release proof-of-concept (PoC) attack code. According to an advisory published by CERT-EU on November 18, no reports of exploitation in real-world environments had been confirmed as of that day.
NGINX fixes 10 vulnerabilities, five of which could allow code execution
The fixes are not limited to the two HTTP/2 flaws. NGINX's November 18 release notes list a total of 10 vulnerabilities. Of these, CERT-EU points to the following five as potentially leading to unauthenticated remote code execution.
| CVE number | Affected component | Main conditions for impact |
|---|---|---|
| CVE-2026-123007 | HTTP/2 flow control | HTTP/2 enabled. Static file delivery is also affected |
| CVE-2026-123633 | HTTP/2 connection management | Reverse proxy setup using HTTP/2 |
| CVE-2026-123634 | DNS name resolution | Proxy destination specified with variables and specific subrequest handling |
| CVE-2026-121810 | HTTP/3 connection management | Reverse proxy setup using HTTP/3 |
| CVE-2026-113425 | Mail proxy authentication | Mail proxy setup using auth_http authentication |
Depending on conditions, all five can lead to worker process crashes or arbitrary code execution. CVE-2026-123007 in particular affects common web server setups with HTTP/2 enabled, so it is a high priority as a problem reaching widely used configurations.
The remaining five include an infinite loop in request body handling under TLS 1.3, a NULL pointer dereference in PROXY protocol v2, a buffer overflow when using alias and try_files, a client certificate verification bypass, and an input validation flaw in the ssl_preread module. Their impact also varies depending on the features and settings in use, so disabling HTTP/2 alone does not address all of the vulnerabilities disclosed this time.
In Japan, Xserver temporarily suspends HTTP/2
As an example of a response in Japan, Xserver announced in a notice dated November 19 that it had disabled HTTP/2 as a provisional countermeasure against these vulnerabilities. The switch was carried out from around 3:00 a.m. to around 4:15 a.m. Japan time and covered six of the company's services.
The affected services are Xserver, XServer Business, XServer Shop, XServer Drive, XServer for WordPress, and XServer Static. The company explained that it investigated the five code-execution-related vulnerabilities and confirmed that three of them, CVE-2026-123634, CVE-2026-121810, and CVE-2026-113425, do not affect it because of how its services are configured. For the HTTP/2-related CVE-2026-123007 and CVE-2026-123633, it took a provisional measure by switching communication to HTTP/1.1.
According to Xserver, users do not need to change any settings, and normal website browsing and various functions remain available. However, depending on the environment and site configuration, display speed may drop slightly, and programs that assume HTTP/2 may stop working properly.
As of its November 19 announcement, the company had not confirmed any attacks or damage exploiting these vulnerabilities. It plans to re-enable HTTP/2 once updates to the fixed versions are complete and to announce the resumption on the same notice page. No specific time for resumption has been given.
Switching to HTTP/1.1 is a measure that stops using the HTTP/2-specific vulnerable code. Because encrypted communication with TLS remains available over HTTP/1.1, there is no need to stop HTTPS.
Even with a CDN, check the HTTP/2 setting on your origin
These vulnerabilities also matter to administrators of websites that use a CDN. A CDN sits between users and the origin server, caching content and relaying traffic, but the HTTP version is determined independently for the connection from the user to the CDN and for the connection from the CDN to the origin.
For example, according to Cloudflare's official documentation, HTTP/2 connections to the origin server are enabled by default on its CDN. If the origin supports HTTP/2, that leg uses HTTP/2; if not, the connection falls back to HTTP/1.1.
In other words, whether HTTP/2 or HTTP/3 is used between the browser and Cloudflare does not reveal whether the origin NGINX is handling HTTP/2. Administrators need to check the CDN's origin connection settings and which HTTP versions the origin server actually accepts.
Wiz Research explains that if a CDN or WAF that terminates HTTP/2 traffic does not pass the traffic needed for an attack through to NGINX unchanged, it may be able to block some attacks. The effectiveness, however, varies by implementation. Where an HTTP/2 connection is established from the CDN to the origin, or where a path remains for connecting directly to the origin NGINX, the impact still needs to be assessed.
On Cloudflare, "HTTP/2 to Origin" can be disabled under "Protocol Optimization" in "Speed." Alternatively, if HTTP/2 is disabled on the origin NGINX, Cloudflare connects over HTTP/1.1. Either can serve as a stopgap, but the permanent fix is to apply the patched NGINX versions.
Fixed versions are 1.30.7 and 1.31.9; watch your OS update status
According to the official NGINX security advisories, the two HTTP/2 flaws were fixed in version 1.30.7 of the NGINX Open Source stable branch and 1.31.9 of the mainline branch.
| Vulnerability | Affected upstream NGINX Open Source versions | Fixed versions |
|---|---|---|
| CVE-2026-123007 | 1.9.5–1.31.8 | 1.30.7 or later / 1.31.9 or later |
| CVE-2026-123633 | 1.19.7–1.31.8 | 1.30.7 or later / 1.31.9 or later |
The commercial NGINX Plus is also affected, so users need to check F5's fix information for the release line they use. Canada's cybersecurity authority also published an alert urging updates on November 18, pointing to F5's individual advisories.
If NGINX was installed through an OS package management system such as Ubuntu or Debian, the update method differs. OS distributors sometimes backport security fixes to an existing version line, so safety cannot be judged from the NGINX version number alone.
As of a check on November 19, 2026, Japan time, Canonical's tracker listed nginx packages on Ubuntu 26.04 LTS and others as affected by CVE-2026-123007. The Debian security tracker likewise classified packages delivered through the security update channels for bookworm and trixie as affected. Administrators should check the OS and package versions they use and compare them against the distributor's fix status.
In environments where the fixed version cannot be applied immediately, one option is to temporarily disable NGINX's HTTP/2 feature. On NGINX 1.25.1 and later, set http2 off; in the relevant server block, and if HTTP/2 is specified in the listen directive, remove that specification. After changing the configuration, check for any effect on operation, reload NGINX, and confirm that the old worker processes have exited.
The Wiz Research technical report also recommends investigating suspicious worker process crashes and unexpected commands launched from NGINX. These are not indicators of compromise specific to these vulnerabilities, but they can offer clues for looking into signs of an attack.
The priority is to identify NGINX instances that are reachable from outside with HTTP/2 enabled and apply the fixed versions. Even static sites cannot be excluded, and in setups that go through a reverse proxy or CDN, you need to check the server that actually handles HTTP/2.
