Critical NGINX HTTP/2 Flaws Allow Remote Code Execution

Two high-severity vulnerabilities in NGINX have been disclosed that could permit unauthenticated attackers to execute arbitrary code on web servers. The bugs, designated CVE-2026-123633 and CVE-2026-123007, were made public on November 18, 2026, and both hinge on the server accepting client-side HTTP/2 connections. Yet they manifest under different server configurations. One arises in reverse proxy setups and the other even in simple static-file servers.

These weaknesses lie within NGINX itself—not the applications it fronts—raising the stakes considerably. Because NGINX is frequently used to handle HTTPS traffic and distribute requests to downstream services, a compromised instance poses a risk to all services behind it. An exploit could expose sensitive data, modify content, or disrupt service altogether.

What the flaws are and who discovered them

CVE-2026-123633 is a use-after-free issue, where an NGINX worker process may reuse memory after it’s already freed. This vulnerability arises when HTTP/2 is enabled from clients and requests are proxied to a backend, particularly under heavy load or when available connections are scarce. Under those circumstances, attackers can trigger crashes or execute code remotely.

Meanwhile, CVE-2026-123007 is significantly more pervasive. It affects NGINX Open Source versions stretching back to 1.9.5 (released in 2015). Notably, it can be leveraged on servers doing nothing beyond serving static content—no backend applications or third-party modules required. Even the most basic web server setups are vulnerable if they expose HTTP/2 to clients.

Affected versions & mitigation steps

Both issues carry a critical rating of 9.2 under CVSS 4.0. Several independent researchers from teams like Ikotas Labs, Metnir, Rapid7, and Xint contributed to the findings.

Fixed versions are now available. For NGINX Open Source, the patches arrived in version 1.31.9 (mainline branch) and 1.30.7 (stable branch). NGINX Plus users are also covered in newer 37.x builds specific to each CVE.

Admins are advised to focus first on servers exposed to the internet, then internal proxies and minimal static sites. It’s critical to confirm the running version on every instance; merely reloading configuration doesn’t replace the vulnerable binary, so service restarts or container replacements are needed.

If updates aren’t yet distributed through your software channel, there’s a temporary workaround: disable client-facing HTTP/2 entirely on listeners where feasible. After that, test externally to ensure HTTP/2 is no longer accepted. But note: disabling the protocol only limits exposure—it doesn’t resolve the root defects.

These vulnerabilities are distinct from earlier NGINX HTTP/2 issues patched in October or a proxy/gRPC bug from June, which depended on different configurations. Having installed the previous patches does not guarantee protection against these new flaws.

What this means: With over 225 million websites running NGINX, the impact could be widespread depending on configuration and version.

From a broader standpoint, these two flaws underscore how complex protocols like HTTP/2 can introduce risk even in simple setups. Organizations should double-check whether HTTP/2 is exposed on any listener and rapidly assess both internal and external servers. The urgency of deploying updates is high—any delay leaves services open to remote takeover or data compromise.