Security firm Xint says it found five remote code execution flaws in nginx, the web server software that Wiz Research says powers more than 225 million websites. Xint’s 18 November post also says nginx’s advisory list recorded 26 new vulnerabilities by 17 November, against two in 2025.
Our number: that is 13 times last year’s count before the five new flaws are added (our arithmetic). The fixed release is nginx 1.31.9, per the nginx security advisories page.
Business Pill · Least Privilege
An evergreen Business Engineer explainer on access control for software. It teaches the general idea only and does not discuss these vulnerabilities.
The key insight: nginx’s advisory list recorded 26 new vulnerabilities by 17 November against two in all of 2025, per Xint, and five research teams then reported critical HTTP/2-dependent flaws found partly with AI tools; the fix is nginx 1.31.9.
Update, 19 November 2026, 06:38 UTC: Google says Wiz used Gemini 4 Carbon to find and validate two of the nginx flaws. In a post on X late on 18 November, Google said “F5 released coordinated patches for two critical remote code execution flaws (CVE-2026-123633 and CVE-2026-123007) in NGINX”. It added that Wiz Research “paired Wiz Atlas with frontier models including Gemini 4 Carbon to uncover and validate the HTTP/2 bugs in parallel with other research teams”, “helping F5 ship fixes before attackers could strike” (Google on X). That matches the two CVEs this piece already lists. Google names Gemini 4 Carbon; it does not say which other models were used.
What Xint and Wiz Disclosed
Xint calls the set “nginx2root”: five nginx flaws plus a separate Linux kernel bug that, it says, can be chained with any of them to gain root on the host. It writes that the most serious, CVE-2026-123007, “allows an unauthenticated attacker to execute commands on an affected nginx server with HTTP/2 enabled”, and that it affects “All official releases from nginx 1.9.5 through 1.31.8”.
Wiz says it, Ikotas Labs, Metnir, Rapid7 and Xint “independently uncovered” two of the flaws, CVE-2026-123633 and CVE-2026-123007, both rated Critical with a CVSS 4.0 score of 9.2, and that both require HTTP/2 to be enabled. We do not reproduce technical exploitation details.

What Was Posted on X
Google’s own post, late on 18 November, says F5 released coordinated patches for two critical remote code execution flaws in NGINX and that Wiz Research paired Wiz Atlas with frontier models including Gemini 4 Carbon to find and validate the HTTP/2 bugs. The update box above quotes it.
How AI Fits In
Xint says a researcher found CVE-2026-123007 while preparing for ZeroDay Cloud 2026, and that “A few days later, a fully automated Xint scan found the other four without human guidance”. It adds: “AI-assisted research is finding bugs faster than ever, even in code as heavily studied as nginx”.
Wiz says it used Wiz Atlas, “the harness powering Wiz’s AI Code scanning solution”, to discover both bugs, and that its researchers “used Gemini 4 Carbon and ChatGPT 5.6 Sol to validate the impact”. Xint’s post names no model in the text we read; it says Xint is a member of OpenAI’s Daybreak cybersecurity program and Anthropic’s Cyber Verification Program.
How Many Servers Are Exposed
Xint writes that as of 10 November, Shodan returned 56.7 million hosts for nginx, that 11.94 million reported an identifiable version, and that “98.9% fell within the version range affected by CVE-2026-123007”. That is about 11.8 million hosts by version string (our arithmetic); Xint cautions the numbers “don’t confirm that a server is vulnerable”, since HTTP/2 must be on and fixes may be backported.
From Report to Patch
Xint lists 16 October 2026 as the date it reported the flaws to F5 through HackerOne and 19 November (Korea Standard Time) as the scheduled patch release; Wiz lists 18 November as F5’s security release. From Xint’s report to the scheduled patch is 34 days (our arithmetic). Xint says that until the update, disabling HTTP/2 on affected ports blocks CVE-2026-123007 and CVE-2026-123633 but not the other three.
Part of a Week of AI-Found Flaws
The disclosure follows Calif’s Java4Shell, found with OpenAI’s Codex, GPT-6 Cyber’s Apple findings, James Forshaw’s Windows UAC bypass and Bitdefender’s report on GitHub traps for AI agents, in a week when officials told Congress patching AI-found flaws drove recent outages.
The Structural Read
As we read the two posts, the bottleneck has moved from finding bugs in heavily reviewed code to shipping and deploying fixes.
As we read Wiz’s account, five teams converging on the same flaws suggests AI-assisted scanning is now common practice at security firms.
Xint, 18 November 2026
“AI-assisted research is finding bugs faster than ever, even in code as heavily studied as nginx”
Three Implications
FOR OPERATORS Xint says to update to 1.31.9 and, until then, disable HTTP/2 on affected ports, which it says blocks two of the five flaws.
FOR PATCH TEAMS Xint warns that Linux distributions often backport fixes without changing the version nginx reports.
FOR THE SECURITY MARKET Both disclosing firms describe AI tooling in the discovery or validation of the flaws.
The Business Engineer Lens
This story maps onto the Business Engineer framework The Agentic Harness War.
The analysis puts it this way: “Foundation models become relatively commoditized as orchestration layers capture differentiation.”
As we read the disclosures, Wiz crediting its own harness, Wiz Atlas, for discovery fits that pattern. The analysis does not discuss these vulnerabilities.
What Is Not Established
The vulnerability details and exposure counts are Xint’s and Wiz’s; both sell security products. We did not test any server, and version strings do not prove a host is vulnerable. We did not see F5’s own advisory text beyond the nginx.org listing. We did not contact Xint, Wiz or F5.
The Bottom Line
Five research teams reported critical nginx flaws found partly with AI tools, as nginx’s advisory count reached 26 this year against two in 2025, per Xint; the fix is nginx 1.31.9.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Vulnerability details, discovery accounts, exposure counts and timelines are from Xint’s and Wiz’s 18 November 2026 posts, read by us; fixed versions are from the nginx.org advisories page. The 13-times ratio, the about 11.8 million hosts and the 34 days are our arithmetic. No exploitation details are reproduced. We did not contact the companies.
Sources: Xint, NGINX2Root (18 Nov 2026) · Wiz Research, Unauthenticated RCE in NGINX (18 Nov 2026) · nginx security advisories









