Stuart Neal
Director of Operations at Nextgen Marketing
We have worked with Vitalli and his team for well over a year now and will continue to do so in the future.
Having the confidence in an agency that allowed us to scale up and down development resources, as and when needed, has really made a positive difference to our agency.
Home / Blog / News / Two critical NGINX HTTP/2 flaws can reach code execution. Updating WordPress will not help
Wordfence put out a public service advisory on 18 November about two critical HTTP/2 vulnerabilities in NGINX, and the framing is worth repeating to clients before they ask the wrong question. Both can lead to remote code execution on the server. Neither is a WordPress problem, neither is fixed by updating WordPress, plugins, themes or a security plugin, and no scan from inside the site can confirm whether your server is patched. That line between what lives in the application and what lives under it is exactly why knowing which stack each client site actually runs on is part of the job rather than a hosting detail.
Fixed open-source versions 1.30.7 (stable) and 1.31.9 (mainline) were released on 18 November 2026. Both flaws require NGINX to accept client HTTP/2. CVE-2026-123633 additionally requires forwarding requests to a backend while a worker runs low on connections.
HTTP/2 is off unless a server’s configuration enables it, and NGINX’s source build omits the HTTP/2 module by default — but NGINX’s own prebuilt packages include it, and hosting configurations vary enormously. That is the whole uncertainty in one sentence: you cannot answer this from the WordPress side.
On the severity question, Wordfence is careful to separate what is proven from what is claimed. Wiz reports lab code execution for both flaws, and Metnir reports it for CVE-2026-123633. Rapid7 published an RCE proof of concept for CVE-2026-123007 against a freshly started local Docker target, needing substantial traffic — which does not rule out other methods. Xint, credited by F5 for CVE-2026-123007, reports code execution over public HTTPS on AWS, GCP and Azure under concurrent normal traffic, and says it chained the NGINX flaw with a separate Linux kernel bug to reach root in testing. CERT-EU reported no known exploitation in the wild as of 18 November.
NGINX either serves the site or sits in front of the application server, and it handles HTTP/2 before anything reaches PHP or WordPress. Successful exploitation runs code with the NGINX worker’s privileges, which puts every site and every database on that server at risk — not just the one that was targeted. On shared or multi-tenant hosting that is the part worth sitting with.
Wordfence makes the limitation of its own product explicit, which is a reasonable thing to pass on to a client: the Wordfence firewall runs at the PHP application layer, and a Wordfence scan cannot confirm the server’s patch status.
One checking detail that catches people out: a site’s public HTTP/2 connection may terminate at a CDN rather than at the NGINX origin. Check the origin itself and any direct route to it, including reverse proxies and containers. And follow the vendor advisory rather than the version string — backported fixes often keep the older NGINX version number, so a site can look unpatched when it is fine, or the reverse.
Most agencies do not know, off the top of their head, which web server each client site runs on. The sites are spread across whatever host the client had when they arrived. The practical move this week is not heroics, it is an inventory: which client sites sit on NGINX, who the host is for each, and what that host has said. For managed hosts the answer is usually “they have handled it or they will” — but it needs to be confirmed per host, not assumed across the portfolio.
If your site is covered by one of our maintenance plans, we are going through the hosting for each site, checking whether NGINX is in the path and whether your provider has shipped a fixed build — and chasing them where they have not. Where the server is under our control the update or the interim mitigation is being applied directly. It will appear in your next monthly maintenance report with what we found and what was done, and there is nothing you need to do in the meantime. If you are not on a plan, this is the sort of thing a plan exists to catch: a critical flaw that is invisible from inside your website, that your own updates cannot fix, and that someone has to go and verify with your host.
Source: Wordfence