1. NGINX vulnerability patched on the day of disclosure
A critical NGINX vulnerability that lets an attacker run code on a server without logging in was disclosed and patched on November 18, 2026. Ikotas Labs, working with OpenAI, named the flaw NGINX2Shell. Tracked as CVE-2026-123007 and scored 9.2 on the CVSS v4.0 scale, it affects NGINX Open Source versions 1.9.5 through 1.31.8, and F5 has released fixed versions 1.31.9 and 1.30.7.
NGINX is among the most widely deployed pieces of web server software in existence, which is why a bug of this kind draws immediate attention. The affected range reaches back to version 1.9.5, a release that dates from 2015. In other words, the faulty code has been shipping in production software for more than a decade without being caught.
The flaw sits in the way NGINX handles HTTP/2, the protocol that most modern browsers and applications use to talk to web servers. A second, related bug in the same part of the code was made public at the same time. Security teams now face a familiar race: the fix is available, but so is detailed knowledge of the weakness, and every unpatched server remains exposed until its operator installs the update.
2. What is the NGINX2Shell vulnerability?
NGINX2Shell is a pre-authentication remote code execution bug in NGINX's HTTP/2 handling. Under the right conditions, someone who can reach an affected server over the network can make it execute code of their choosing, with no password, no account and no prior access of any kind.
The two halves of that description explain the severity rating. Remote code execution means the attacker is not limited to reading data or crashing a service; they can run their own instructions on the machine. Pre-authentication means there is no login step standing in the way. Bugs that combine both properties are treated as the most serious class of server vulnerability, because the usual first line of defense, proving who you are, never comes into play.
The findings indicate that the flaw may also be used to cause a denial of service, knocking a server offline rather than taking control of it. In both cases the precondition is that HTTP/2 is switched on. A server that does not accept HTTP/2 connections is outside the reach of this particular bug.
The identifier CVE-2026-123007 places the issue in the public catalog used across the industry to track security flaws, and the 9.2 score under version 4.0 of the Common Vulnerability Scoring System puts it in the Critical band, the highest of the scale's severity categories. The vulnerability was identified in NGINX version 1.31.8, and the affected range covers NGINX Open Source releases from 1.9.5 up to that version under specific conditions.
3. A second HTTP/2 bug disclosed alongside it
Researchers also published details of CVE-2026-123633, a separate weakness in the same HTTP/2 implementation. It is a use-after-free issue, a category of memory error in which software continues to use a region of memory after it has been released for other purposes. Flaws of that type can allow an attacker to corrupt data or redirect a program's behavior.
This second bug has a smaller footprint. It affects NGINX Open Source versions 1.19.7 through 1.31.8, a shorter span than the main flaw, and exploiting it requires additional conditions to be met. Those extra requirements narrow the set of servers that could realistically be attacked through it.
Both vulnerabilities became public on the same day, and both are closed by the same updates. For administrators, that simplifies the response: a single upgrade addresses the pair, and there is no need to assess them one at a time before acting.
The pairing also says something about where the weakness lay. HTTP/2 is considerably more intricate than the protocol version it succeeded: it carries many requests over a single connection and keeps track of each one separately, which demands careful bookkeeping of memory and state. Code of that kind offers more opportunities for subtle mistakes, and two distinct flaws surfacing in the same component suggests it had not received scrutiny in proportion to its exposure.
4. F5's fixes and what administrators should do
F5, the company that owns NGINX, issued patched releases on November 18, the day the flaws were disclosed. The corrected versions of NGINX Open Source are 1.31.9 and 1.30.7, and upgrading to one of them is the immediate step for anyone running an affected version.
Publishing a fix at the moment of disclosure is the standard goal of coordinated vulnerability handling. Researchers report a problem privately, the vendor prepares an update, and the details are made public only when a remedy is ready. That sequence gives defenders a way to protect themselves from the first hour, but it also starts a clock. Once a patch exists, anyone can study what changed and work out what the original weakness was.
For that reason the period between disclosure and widespread patching is the main source of risk. Organizations that delay the upgrade, or whose update fails, remain exposed to a flaw whose existence and general nature are now public knowledge, and the findings warn that such delays could result in severe breaches.
Teams that cannot update right away are advised to establish two things: whether HTTP/2 is enabled on their NGINX instances, and how exposed those instances are to outside traffic. Because the bug depends on HTTP/2, a server's configuration determines whether it is reachable through this route at all, and an internet-facing machine carries far more risk than one that sits behind other controls.
5. Proof-of-concept work and the state of exploitation
Rapid7 has shown that the flaw can be turned into a functioning attack. The security firm demonstrated a working proof-of-concept exploit against an official NGINX 1.31.8 Docker image, the packaged form in which many organizations deploy the software.
The demonstration came with a practical caveat. Rapid7's attack needed roughly 128 gigabytes of traffic under laboratory conditions. That volume does not make exploitation impossible, but it means an attack is noisy and takes sustained effort rather than a single crafted request.
As of the disclosure, no active exploitation in the wild had been reported. That status can change quickly once the mechanics of a vulnerability are understood, which is why the absence of known attacks is not treated as a reason to postpone patching. The existence of a working proof of concept shows that the path from theoretical flaw to practical attack has already been walked at least once.
For defenders, the figure offers a limited kind of reassurance. An attack that depends on pushing that much data at a target is more likely to register in traffic monitoring than one hidden in a handful of packets. It should not be mistaken for protection, though. Laboratory demonstrations establish that an attack is possible, not that it is the most efficient version, and techniques tend to improve once more researchers and attackers begin studying the same flaw.
6. Who found the flaw, and what was OpenAI's role?
Ikotas Labs disclosed NGINX2Shell and, in a post on X dated November 18, credited OpenAI for collaborating on both the discovery and the disclosure of the bug. The post did not turn the episode into a single-team story, however, because Ikotas was not alone in spotting problems in this part of NGINX.
Overlapping vulnerabilities were identified independently by Wiz, Rapid7 and Xint. Several groups arriving at the same weaknesses within the same window is unusual for code that had gone unexamined, or at least unreported, for so many years.
That convergence is being read as a sign of how vulnerability research is changing. AI tools are increasingly part of how researchers search large codebases for subtle memory and logic errors, and a bug that survived more than ten years of human review was found by multiple teams in close succession. The credit given to OpenAI in a coordinated disclosure of a flaw rated 9.2 offers a concrete, checkable example of an AI developer contributing to defensive security work.
AI laboratories have an obvious interest in such examples, since they demonstrate usefulness in a field where the stakes are easy to understand. The same capability raises a harder question for defenders. If automated tools make long-hidden flaws easier to find, they do so for anyone who uses them, and the advantage goes to whichever side looks first.
7. Why a web server bug has such wide reach
NGINX is used both as a web server and as a reverse proxy, and it handles a significant share of global internet traffic. In the reverse proxy role, it sits in front of other servers, receives requests from the outside world and passes them along to the systems behind it.
That position is what makes a compromise so damaging. A reverse proxy is the front door to everything it protects: application servers, databases and internal services that are never meant to be reached directly. An attacker who gains code execution on the proxy is placed at the point through which all of that traffic flows, with the ability to observe it, alter it or use the foothold to move further into a network.
The software's ubiquity extends to online financial services, including the websites and interfaces of trading platforms and other businesses that handle customer funds, which commonly place a proxy layer between users and their core systems. For those operators the guidance is the same as for anyone else running an affected version: move to 1.31.9 or 1.30.7, and until that is done, confirm whether HTTP/2 is enabled and how much of the deployment faces the open internet.
