Two vulnerabilities make it possible to remotely take control of NGINX servers, the software that runs a large part of the web.
One of them had been sleeping in the code since 2015.
On November 18, 2026, F5, the company that publishes NGINX, released a security update correcting 10 vulnerabilities.
Two of them, referenced CVE-2026-123633 and CVE-2026-123007, received a score of 9.2 out of 10 and allow an attacker to execute code on the targeted server without having to identify themselves.
Five research teams discovered them each on their own, including that of Wiz, who says he found them using his AI code analysis tool.
Black Friday: your digital peace of mind at -61%
Secure your purchases during Black Friday with the best offer of the year. Find the good deals, Bitdefender takes care of the scams.
Software placed in front of millions of sites
If the alert is so critical, it is primarily because of the targeted software. When an Internet user visits a site, their request often goes first through NGINX. This open source web server receives traffic, manages HTTPS encryption and redirects each request to the correct application.
Wiz estimates that it powers over 225 million sites. As a single installation often serves several services, taking control of it allows you to steal data, modify pages or cut everything, even if the applications behind are perfectly secure.
Both flaws only affect servers where HTTP/2 is enabled. This very widespread version of the web protocol sends several requests in parallel over the same connection.
Two errors in memory management
The first flaw, CVE-2026-123633, is a “ use after release “. NGINX keeps a list of connections that it can close to make room. However, a connection could remain there while a request was still in progress. NGINX thus risked closing it and freeing its memory, which the query continued to use. An attacker capable of slipping his own data into it can then hijack the program. This flaw, present for almost 6 years, only concerns servers which relay requests to another application.
The second, CVE-2026-123007, has a broader scope. NGINX keeps a counter of the amount of data it can send to a client. When a request is canceled, it returns the unused credit to this counter without checking that the total does not exceed the authorized maximum. The counter may then display an outlier, which ultimately causes data to be copied outside of the intended memory area. Wiz says it got code running on a simple static file server, and the flaw dates back to version 1.9.5, released in 2015.
An 11-year-old flaw
Wiz researchers explain that they spotted the two vulnerabilities with Wiz Atlas, their AI code analysis platform, which combines several models. Finding this kind of flaw requires tracking subtle interactions between numerous execution paths, a job long reserved for a few specialists. “Complexity is no longer an obstacle to discovering or exploiting these vulnerabilities », They write, recalling the precedent of PHP2Shell, two critical PHP flaws corrected in October.
Wiz has not published his exploits. The researchers point out that these flaws remain difficult to exploit, because the attack must be adapted to each version and each configuration of NGINX. In May, another critical flaw in NGINX was exploited just a few days after the release of its patch.
Administrators should upgrade to NGINX 1.30.7 or 1.31.9. If they can’t do this right away, Wiz advises disabling HTTP/2 temporarily.
All the tech news at a glance
Add Numerama to your home screen and stay connected to the future!

