NGINX serves as the front‑door for millions of web applications, making its HTTP/2 flaws a systemic risk.
*Critical unauthenticated remote code execution flaws in NGINX's HTTP/2 stack threaten over 225 million websites. Two CVEs, both rated 9.2, demand immediate patching or HTTP/2 disablement.*
Two critical remote code execution bugs have surfaced in NGINX's HTTP/2 handling, threatening the backbone of the modern web. The flaws—CVE‑2026‑123633 and CVE‑2026‑123007—allow unauthenticated attackers to run arbitrary code on any server that has HTTP/2 enabled. With NGINX serving as the default front‑door for more than 225 million sites, the potential fallout spans from data theft to full service disruption. Independent researchers from Wiz, Ikotas Labs, Metnir, Rapid7, and Xint uncovered the bugs simultaneously, confirming their severity and the urgent need for remediation.
Two critical vulnerabilities, CVE‑2026‑123633 and CVE‑2026‑123007, affect NGINX when HTTP/2 is enabled. Both are unauthenticated remote code execution bugs rated 9.2 on the CVSS scale. CVE‑2026‑123007 traces back to NGINX 1.9.5 released in 2015, meaning the integer underflow has hidden for more than a decade. CVE‑2026‑123633 is a use‑after‑free flaw present for nearly six years. Exploitation lets an attacker execute arbitrary code, steal data, tamper with sites, or disrupt services. The bugs reside in NGINX itself, so even perfectly coded back‑end applications remain exposed. Because NGINX is the default front‑door for modern web architectures, a single compromised instance can jeopardize multiple applications and APIs.
Five independent research groups—Wiz Research, Ikotas Labs, Metnir, Rapid7, and Xint—converged on the same flaws. Wiz used its Atlas platform, powered by AI code‑scanning, alongside Gemini 4 Carbon and ChatGPT 5.6 Sol to trace subtle memory‑safety interactions. Their tests showed code execution on an NGINX server serving static files, with no backend or third‑party modules required. The discovery underscores how AI models now automate the tracing of complex memory operations that previously formed a practical barrier to exploitation.
Wiz reported the first use‑after‑free issue on October 21 2026. F5 supplied a patch on November 10 2026 for CVE‑2026‑123633. The second integer‑underflow bug was reported on November 16 2026, confirmed on November 17, and included in F5’s coordinated public release on November 18 2026. Organizations must inventory NGINX deployments, verify HTTP/2 is enabled, and apply the November 18 security update immediately. Where patching is delayed, disabling HTTP/2 provides a temporary mitigation.
NGINX powers more than 225 million websites, acting as the gateway for countless APIs and services. A single vulnerable front‑end can expose multiple downstream applications, magnifying the attack surface. The fact that AI accelerated the discovery of flaws hidden for years signals a shift: complexity can no longer be relied upon as a defense. Faster vulnerability discovery shortens the window between disclosure and exploitation, intensifying the workload for security teams tasked with patch validation and rollout.
All versions dating back to NGINX 1.9.5 released in 2015 are vulnerable to CVE‑2026‑123007.
Both CVE‑2026‑123633 and CVE‑2026‑123007 require HTTP/2 to be enabled on the affected NGINX server.
If patching cannot be done immediately, disabling HTTP/2 on the NGINX server provides a temporary mitigation.
The clock is already ticking. Patches landed on November 18 2026, but many operators still run legacy versions with HTTP/2 enabled. Immediate identification, patching, or temporary HTTP/2 disablement is the only viable defense. As AI continues to lower the barrier to finding deep memory‑safety bugs, the security community must accelerate response cycles or risk repeated, large‑scale compromises across the internet's most ubiquitous web server.
Sources: Hacker News article, https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities