F5 Patches Five NGINX Vulnerabilities With Working Exploits
$ grep -o 'CVE-[0-9-]*'
┌─ f5-patches-critical-nginx-flaws.md :: hero
F5 released fixes on Tuesday for five newly disclosed vulnerabilities in NGINX, each accompanied by a working exploit. The most severe flaw, tracked as CVE-2026-123007 with a critical CVSS score of 9.2, allows unauthenticated attackers to run code remotely over the internet and was chained by researchers with a Linux kernel flaw to gain root access. Because NGINX is widely deployed and handles traffic for roughly a third of all websites, organisations and website operators worldwide, including in Malaysia, face potential exposure unless affected servers are updated with F5’s official patches.
Key Facts
| Attribute | Value |
|---|---|
| Affected product and versions | NGINX (exploit demonstrated on NGINX 1.31.8; F5 BIG-IP and F5OS not vulnerable; F5 cloud services under evaluation) |
| CVE IDs | CVE-2026-123007 (four other flaws also disclosed) |
| CVSS | 9.2 |
How Does CVE-2026-123007 Work?
CVE-2026-123007 is an integer underflow in NGINX’s HTTP/2 implementation that lets an unauthenticated attacker trigger an out-of-bounds memory copy and run code on the server. Four additional vulnerabilities affect reverse proxying over HTTP/2 and HTTP/3, combinations of the resolver and mirror directives, and mail proxying.
In HTTP/2, clients grant servers permission to send a set volume of data through a credit system. When a client cancels a request while response data is still queued, NGINX returns the unused credit without verifying whether the total exceeds the protocol’s maximum. A subsequent check subtracts that inflated figure from the maximum using unsigned arithmetic, causing the calculation to wrap into an enormous value treated as a negative response length. That creates a memory buffer whose end pointer precedes its start pointer, resulting in an out-of-bounds memory copy. Separately, Wiz Research, working with Ikotas Labs, Metnir, and Rapid7, documented a use-after-free bug in reverse-proxy configurations when connection slots run low.
CVE-2026-123007 is a critical HTTP/2 integer underflow scored at 9.2 on the CVSS scale that lets an unauthenticated attacker run code on a server from anywhere on the internet.
How Practical Are the Working Exploits?
Every one of the five disclosed vulnerabilities has a working exploit, though exploiting CVE-2026-123007 with the known technique carries strict network and hardware conditions. Researchers also demonstrated that the HTTP/2 bug can be chained with a separate Linux kernel flaw to gain root privileges on the host machine.
Rapid7 researcher Stephen Fewer built a working exploit against an NGINX 1.31.8 container configured with HTTP/2 and TLS. The attack requires roughly 128 gigabytes of HTTP/2 traffic, mostly empty control frames. In testing, it completed in two-and-a-half minutes on a fast local network and would take around two minutes on a 10-gigabit link or 17 minutes over 1 gigabit. Because NGINX closes connections after one hour by default, networks slower than roughly 300 megabits per second cannot finish in time. The method also relies on a specific optimised memory-copy routine in the GNU C Library selected on x86-64 processors with certain AVX-512 features, though Rapid7 noted other unknown methods may exist. Separately, Xint chained the HTTP/2 flaw with a Linux kernel SCTP networking vulnerability to reach root.
Exploiting CVE-2026-123007 via the known technique requires about 128 gigabytes of HTTP/2 traffic and cannot complete within NGINX’s default one-hour connection timeout on links slower than roughly 300 megabits per second.
How Were the Flaws Found and Who Is Affected?
Researchers at Rapid7, Wiz Research, Xint.io, Ikotas Labs, and Metnir uncovered the vulnerabilities, with Xint and Wiz using automated harnesses and AI models. A Shodan snapshot by Xint showed 98.9% of 11.94 million version-identifiable NGINX hosts running affected releases, while F5’s cloud services remain under evaluation.
Xint pointed an automated testing harness at roughly 50,000 lines of NGINX code without human guidance, uncovering four of the five flaws; high-severity findings emerged in 90 minutes, with a working exploit following 10 minutes later. Wiz Research independently found two of the flaws and validated its findings using AI models, attributing the convergence of multiple teams on long-hidden bugs to AI’s improving ability to trace memory-safety flaws. F5 confirmed that BIG-IP and F5OS products are not vulnerable.
“When we used Xint to scan NGINX, we didn’t know what to expect, since it’s a mature, widely deployed codebase that’s undergone years of scrutiny. Others have targeted NGINX with AI pentesting, but none found this critical bug, which is a testament to our team and the engineering expertise that’s gone into our AI harnesses. We’re grateful for the collaboration with project maintainers to get a patch released quickly.”
Brian Pak, co-founder and CEO of Xint
Xint’s Shodan snapshot identified 98.9% of 11.94 million version-identifiable NGINX hosts running affected releases, though that version count does not verify HTTP/2 exposure or rule out backported patches.
What Should You Do?
- Apply F5’s official NGINX fixes released on Tuesday, or verify whether your software distribution has supplied a backported patch.
- Audit NGINX configurations for exposure across HTTP/2, reverse proxying over HTTP/2 and HTTP/3, combined resolver and mirror directives, and mail proxying.
- Ensure underlying Linux kernel updates are applied to mitigate privilege-escalation chaining such as the SCTP networking flaw.
- Monitor F5’s ongoing evaluation if your organisation relies on F5 cloud services.
- Malaysian organisations that detect suspicious server activity or suspected compromise can report incidents to Cyber999 (MyCERT).
Common Questions
Are F5 BIG-IP and F5OS affected by these NGINX vulnerabilities?
No. According to CyberScoop, F5 found that its BIG-IP and F5OS products are not vulnerable to these flaws. However, F5’s cloud services are still being evaluated, so organisations relying on them should watch for F5’s status updates while patching affected NGINX deployments.
Does running an affected NGINX version mean a server is definitely exploitable?
Not necessarily. Xint noted that its Shodan scan counted version numbers rather than confirmed exposure, as it cannot verify whether a server accepts HTTP/2 or carries a backported fix. Known exploitation of CVE-2026-123007 also depends on specific network speeds and CPU features.
Why did multiple research teams uncover these NGINX bugs at the same time?
Wiz Research attributed the independent convergence of several teams on the same long-hidden NGINX bugs to artificial intelligence improving at tracing memory-safety flaws. Xint similarly used an automated testing harness without human guidance to scan around 50,000 lines of code and uncover four flaws.
Sources and Methodology
This dispatch is based exclusively on reporting by CyberScoop. No external vendor advisories, code repositories, or third-party documents were independently reviewed or verified.
$ ls -t news/ | head -5
- AI Overreliance Could Erode Critical Thinking for Security Teams
- Google Opens Carbon Access to Paying Developers and Businesses
- Google DeepMind AI Agents Cut Cost to Factor 1024-Bit RSA Keys
- Claude Code Exposes Developer Passwords and Keys on GitHub
- Hackers Use Fake Events to Steal Microsoft and Google Accounts
$ ls ./tags
- #Vulnerabilities
- #AI Security