CVE-2026-123007 — an integer underflow in NGINX’s HTTP/2 handling — has been rated critical (9.2/10 CVSS) and carries a working exploit that, in tests, let researchers run code on a server and, when chained with another bug, gain full root on the host.
CVE-2026-123007: integer underflow in HTTP/2
F5, the company that maintains NGINX, released fixes Tuesday and described the most severe flaw as an integer underflow tracked as CVE-2026-123007. The bug sits in NGINX’s HTTP/2 implementation. HTTP/2 uses a credit system in which a client grants the server permission to send a bounded amount of data; when a client cancels a request while response data remains queued, NGINX returns the unused credit without checking whether the total exceeds the protocol maximum. A later unsigned subtraction wraps the value into an enormous number, which is treated as a negative response length. That produces a buffer whose end pointer sits before its start and permits a memory copy that spills outside the buffer’s boundaries.
Exploit demonstration against NGINX 1.31.8 and practical limits
Rapid7 researcher Stephen Fewer built a working exploit against an NGINX 1.31.8 container with HTTP/2 and TLS enabled. The exploit requires roughly 128 gigabytes of HTTP/2 traffic, most of it control frames that carry no actual data. In a fast local network the demonstration completed in two-and-a-half minutes; the researchers estimated about two minutes over a 10‑gigabit link and 17 minutes over a 1‑gigabit link. Because NGINX closes connections after an hour by default, the team calculated that networks slower than about 300 megabits per second could not complete the attack before the connection closed.
The exploit also depends on a particular optimized memory-copy routine in the GNU C Library, chosen on x86‑64 chips with specific AVX‑512 features. Rapid7 noted that other exploit techniques may exist but said none are currently known.

Your scanner finds 4,000 vulns. Which 12 matter?
Nubivance is a Rapid7 Registered Partner delivering vulnerability management as a service - scanning, risk-based prioritization, and remediation follow-through across IT and OT.
Fix the backlogMultiple teams and multiple bugs: Rapid7, Wiz, Xint, Ikotas and Metnir
F5 credited Rapid7 along with Wiz Research, Xint.io and Ikotas Labs for reporting the flaws. Wiz Research, working with Ikotas Labs, Metnir and Rapid7, independently found two of the same flaws and documented a use‑after‑free bug that can appear in reverse‑proxy setups when connection slots run low. Xint reported four of the five flaws after running an automated testing harness.
Xint’s account is striking: the harness was pointed at roughly 50,000 lines of NGINX code, and within 90 minutes it produced high‑severity findings; a working exploit followed within 10 minutes. Xint’s team then chained the HTTP/2 bug with a separate flaw in the Linux kernel’s SCTP networking code to achieve root on the host. Brian Pak, co‑founder and CEO of Xint, told CyberScoop, “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.”
Other affected NGINX features and scope indicators
The five disclosed vulnerabilities affect multiple NGINX features beyond HTTP/2: reverse proxying over HTTP/2 and HTTP/3, combinations of the resolver and mirror directives, and mail proxying. F5 said its cloud services are still being evaluated for exposure, and that BIG‑IP and F5OS products were not vulnerable.
Xint’s Shodan snapshot found that 98.9% of 11.94 million version‑identifiable NGINX hosts were running affected releases — a metric that counts version strings but does not confirm whether each server accepts HTTP/2 traffic or has received backported fixes.
What this means for technologists, procurement leaders, and threat actors
- Technologists and security teams: verify that NGINX servers are updated to the patched releases F5 published Tuesday; confirm whether exposed hosts accept HTTP/2 and whether connection timeouts or library selections (glibc with AVX‑512 optimizations) could allow exploitation.
- Procurement and enterprise infrastructure leaders: note that BIG‑IP and F5OS were assessed as not vulnerable and that F5 is still evaluating cloud services; inventories that rely on version strings (for example, Shodan snapshots) may overstate immediate exposure and require per‑host confirmation.
- Adversaries and threat actors: the researchers’ demonstrations show an available path to remote code execution and, in at least one chain, privilege escalation to root; Rapid7 warned other exploit methods may exist even if none are currently known.
These disclosures arrive amid a flurry of critical defects in foundational software: researchers disclosed a critical flaw in Java less than 24 hours earlier, and the prior month produced a string of similar vulnerabilities that some described as “Black October.” For now, F5’s fixes are the immediate control point — and F5’s ongoing cloud assessment, the specific conditions required for the exploit, and the rapid independent discoveries by multiple research teams leave a concrete, testable task for operators: patch promptly, inventory HTTP/2 exposure, and confirm whether the specific library and hardware conditions that made the exploit practical exist in their environments.
Source: CyberScoop — It’s NGINX’s turn to cause vulnpocalypse headaches




