Five Critical NGINX Flaws Include 11-Year-Old HTTP/2 RCE Bug

Researchers disclosed five critical NGINX flaws, including an HTTP/2 remote code execution bug dating to 2015; F5 shipped fixes in NGINX 1.30.7 and 1.31.9.
Table of Contents
    Add a header to begin generating the table of contents

    Researchers from five security teams disclosed five critical vulnerabilities in NGINX on November 18, including an unauthenticated remote code execution flaw in HTTP/2 handling that has existed since the protocol was added to the server in 2015. F5 released fixes the same day for a web server that Xint says is used by 88.6% of Fortune 500 companies.

    Five Teams Report Critical Bugs in the Current NGINX Release

    Wiz, Ikotas Labs, Metnir, Rapid7 and Xint independently reported flaws affecting NGINX 1.31.8, the current mainline release. All five CVEs — CVE-2026-113425, CVE-2026-121810, CVE-2026-123007, CVE-2026-123633 and CVE-2026-123634 — carry a critical CVSS 4.0 score of 9.2.

    Xint documented all five under the name “NGINX2Root,” a label that also covers a Linux kernel SCTP bug used to reach root privileges. Ikotas Labs highlighted one flaw it calls “NGINX2Shell,” while Wiz published technical detail on two of the flaws. NGINX powers more than 225 million websites.

    How CVE-2026-123007 Corrupts Memory Through HTTP/2 Flow Control

    CVE-2026-123007 is an integer underflow in HTTP/2 flow control that produces a heap buffer overflow. An unauthenticated remote attacker can corrupt memory in an NGINX worker process, leak memory addresses, and overwrite a cleanup-function pointer to run commands with the worker’s privileges. The affected range runs from NGINX 1.9.5, the release that introduced HTTP/2, through 1.31.8.

    The flaw only matters on servers with HTTP/2 enabled. That condition is why disabling HTTP/2 is the available workaround until a patched build is installed.

    Use-After-Free in CVE-2026-123633 Also Allows Unauthenticated Code Execution

    CVE-2026-123633 is a use-after-free that can likewise give an unauthenticated remote attacker code execution. Wiz says the bug dates to 2020, meaning it has been present in NGINX releases for roughly six years.

    Why Root Access Requires the Linux Kernel SCTP Flaw

    The NGINX bugs on their own run code at worker-process privilege. Reaching root requires chaining them with the Linux kernel SCTP vulnerability. Xint tested that chain on AWS, GCP and Azure, and Luxembourg’s CIRCL confirmed the kernel bug is necessary for root takeover.

    An AI System Found the Bug in About 90 Minutes

    Xint says an AI system flagged the HTTP/2 flaw roughly 90 minutes after being pointed at about 50,000 lines of NGINX code. Some of the teams involved said they collaborated with OpenAI. Xint noted that the last well-known pre-authentication remote code execution bug in NGINX dates to 2013.

    A bug that sat in a widely deployed server for more than a decade surfaced in a single sitting.

    Scale of Exposure Across NGINX Deployments

    Censys researcher Zakir Durumeric says nearly 10 million internet-facing services run vulnerable NGINX versions. A Shodan scan on November 10 found 98.9% of NGINX servers with an identifiable version running a vulnerable release. That figure overstates real exposure: not every one of those servers has HTTP/2 enabled, and not all are unpatched by their vendors.

    No exploitation in the wild has been reported, and none of the flaws appear in CISA’s Known Exploited Vulnerabilities catalog at the time of writing. Xint cautioned that the age of some bugs means earlier exploitation cannot be ruled out.

    F5 Patches in NGINX 1.30.7 and 1.31.9 Alongside 22 Advisories

    The fixes ship in NGINX Open Source 1.30.7 on the stable branch and 1.31.9 on the mainline branch. F5 published 22 advisories on November 18, ten of them covering NGINX. Administrators are directed to read each advisory for configuration-specific mitigations, and disabling HTTP/2 is the workaround while patching is pending.

    The disclosure arrives with an unusually broad population of vulnerable hosts and a public chain that reaches root on major cloud platforms. How quickly operators update, and whether attackers produce working exploits from the published research, will determine whether the absence of in-the-wild attacks holds.

    Related Posts