AI found more critical bugs in another widely used technology running 225M websites
Hot on the heels of serious vulnerabilities in seemingly every other widely used software powering the entire internet, F5 today released emergency patches for 10 bugs in open source NGINX.
These include critical remote code execution flaws that let attackers hijack NGINX servers with HTTP/2 enabled, as well as the millions of websites running NGINX, to steal sensitive data and disrupt services.
As with all of the other recently disclosed vulnerabilities in foundational technologies like PHP, Apache, and Java, these critical NGINX bugs were uncovered by researchers using AI. One of them has been hiding in the code for at least 10 years.
REG AD
Tomer Schwartz, Wiz’s chief technology officer of exposure management and AppSec, helped find the bug, along with researchers from Ikotas Labs, Metnir, Rapid7, and Xint working independently.
REG AD
Schwartz told The Register that this wave of AI-discovered flaws “point to a new reality.”
“We need to treat every internet-connected system as potentially vulnerable,” Schwartz said. “AI has changed not only which vulnerabilities can be found, but how quickly longstanding risks can become real threats.”
NGINX is the world’s most widely used open source web server, handling HTTPS requests and routing incoming traffic to the right application or API. It powers about a third of all websites (around 225 million), and F5 maintains the open source version.
On Wednesday, F5 published an out-of-band security notification that addressed 10 CVEs in open source NGINIX. If you run NGINX, update to 1.31.9 as soon as possible.
CVE-2026-123633 and CVE-2026-123007
Two of the vulns are especially serious as they allow a remote attacker to execute code on the NGINX server without authentication. These two, tracked as CVE-2026-123633 and CVE-2026-123007, both received a critical, 9.2 CVSS rating, and CVE-2026-123007 affects versions dating back to NGINX 1.9.5 in 2015. Both vulnerabilities require HTTP/2 to be enabled on the affected NGINX server.
All of the security shops listed earlier independently found at least one of these two CVEs - most, if not all, with an AI assist - and reported them to F5.
“Both vulnerabilities are memory-safety issues in how NGINX handles HTTP/2 traffic,” Schwartz and fellow Wiz kid Nir Ohfeld said in a Wednesday blog about the two bugs. “In CVE-2026-123633, a connection's memory is freed while a request still needs it, creating a use-after-free. In CVE-2026-123007, incorrect flow-control accounting leads to an invalid memory copy.”
REG AD
Rapid7’s team also published a proof-of-concept to demonstrate leveraging CVE-2026-123007 for unauthenticated remote code execution against a vulnerable linux/amd64 Docker image running nginx version 1.31.8.
Xint found three other NGINX flaws, plus a separate vulnerability in the Linux kernel that can be chained with any of the five NGINX vulnerabilities to gain root privileges on the host. The research team dubbed the six bugs “nginx2root.”
The AI-powered offensive security firm also published videos demonstrating attacks using CVE-2026-123007 to execute commands as the NGINX worker inside a container, and then using the kernel vulnerability to escalate to root on the host. These attacks work on AWS, Microsoft Azure, Google Cloud Platform, and bare-metal servers.
AI-powered bug hunting for the win?
"While preparing for ZeroDay Cloud 2026, a Xint researcher found CVE-2026-123007," the company said in a Wednesday blog. "A few days later, a fully automated Xint scan found the other four without human guidance."
All of these newly-disclosed CVEs point to a larger security issue - especially in light of Tuesday’s Java4Shell, found by AI-bug-hunting-outfit Calif, PHP2Shell (Calif was one of the firms that also used AI to find and weaponize these critical vulns in PHP), and a critical Apache flaw under active exploitation .
“AI is also accelerating vulnerability discovery faster than existing infrastructure was designed to absorb,” Schwartz told us. “Even after a vendor ships a fix, security teams still face heavy manual work in terms of locating vulnerable versions and configurations, testing patches, deploying interim mitigations (such as disabling HTTP/2), and coordinating all of these rollouts.”
There is light at the end of the tunnel, he believes. “In the long run, AI-powered vulnerability research will harden foundational software.”
REG AD
But in the meantime, it’s a nightmare for security teams, vulnerability managers, and system administrators, who are drowning in the near-daily deluge of CVEs to patch before attackers find and exploit these vulnerabilities.
“Continued collaboration between security researchers, maintainers, and defenders is crucial to absorb the speed and scale of these disclosures,” Schwartz said.®
KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.