Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days Business IEA Says Data Center Power Demand Could Double to 1,000 TWh by 2030, With AI Supplying Three Quarters of the Growth Chatbots ChatGPT Now Transcribes as You Talk: OpenAI Removes the 10-Minute Cap and Streams Words Live on Every Plan Business Ex-Alibaba Executive's Everful Raises $48.5 Million to Put AI Agents in Charge of Sourcing From 30,000 Factories AI Agents Google Built a Mirror Dimension of the Internet to Train Its Chrome AI Agents Without Breaking the Real Web Productivity Canva Finally Adds an Off Switch for Its Generative AI Features, and a Filter to Hide AI-Made Templates Business Anthropic Begins Trading Today: 980 Million Shares at $83, a $1.5 Trillion Valuation and the Largest IPO on Record News Google Books Blue Origin New Glenn to Launch TPU-Carrying AI Satellites in Early 2029 Security Hackers Used the AI Penetration Tool ARTEX to Disrupt European Commission Systems for Over Ten Days

Four of Five Critical Nginx Flaws Were Found by an Autonomous AI Agent With No Human Steering

Four of Five Critical Nginx Flaws Were Found by an Autonomous AI Agent With No Human Steering

Security firm Xint disclosed nginx2root: five independent nginx vulnerabilities, each CVSS 4.0 9.2, each enabling unauthenticated remote code execution. The worst, CVE-2026-123007, affects every release since HTTP/2 landed in 2015 and covers 98.9 percent of identifiable-version hosts. Four of the five flaws came from an autonomous AI run. F5 shipped fixes the same day.

Nginx runs roughly a third of all public websites, and for two decades security researchers have fuzzed, audited and hammered it looking for the kind of bug that matters most: a way to execute code on the server without logging in. On November 18, the offensive security firm Xint disclosed nginx2root, a collection of five such vulnerabilities — each independently exploitable for unauthenticated remote code execution and each rated 9.2 under CVSS 4.0 — and the story of how they were found says as much about the state of AI-driven offensive security as the bugs themselves.

The most severe flaw, CVE-2026-123007, sits in the HTTP/2 flow-control code and has been present in every official release since 1.9.5, the 2015 version that introduced HTTP/2. By flooding the server with WINDOW_UPDATE frames, an attacker can push an internal send-window counter past its maximum; a subsequent unsigned-integer underflow turns that corrupted value into an oversized memory copy that discloses addresses, corrupts the heap and ultimately replaces the cleanup function that runs when a connection closes. Wiz Research, one of four other teams that independently found the same or related flaws on the same day, demonstrated working code execution against a stock nginx server serving plain static files over HTTPS — no backend application, no third-party modules.

The exposure is enormous. Shodan data from November 10 cited by Xint counted 56.7 million hosts running nginx, of which 11.94 million reported an identifiable version — and 98.9 percent of those fell within the affected range. F5, which maintains nginx, shipped fixed versions 1.31.9 and 1.30.7 the same day the flaws were disclosed, in coordination with AWS, Azure and Google Cloud. Xint said it had no evidence of exploitation in the wild at publication, while cautioning that some of the bugs have existed for so long that prior abuse cannot be ruled out. A proof-of-concept exploit for the worst flaw, written by Rapid7 researcher Stephen Fewer, is already public on GitHub.

The discovery story is the headline within the headline. While preparing for its ZeroDay Cloud event, an Xint researcher pointed the platform at a 50,000-line slice of the nginx codebase; ninety minutes later it flagged high-severity issues with reproduction steps specific enough to become a working exploit in under ten minutes. Four of the five vulnerabilities came out of that fully autonomous run with no human steering. The company noted that six months ago its CopyFail discovery, the most severe Linux vulnerability in years, still required human guidance — while nginx2root was found using mostly the same models that have been generally available since May. Several of the disclosing teams have said publicly that they used AI models, including models from OpenAI, in their research.

The bugs also chain. Xint paired one of the nginx flaws with a separate vulnerability in the Linux kernel SCTP subsystem to escalate from nginx code execution to full root on the host, escaping a Docker container running nginx in a demonstration. In practice, an externally reachable TLS/HTTP/2 port is enough to fingerprint a target, corrupt the send window and start issuing commands with worker privileges — and because nginx so often terminates TLS in front of internal services, one compromised proxy can expose everything behind it.

The disclosure process itself is being rebuilt around the new threat model. F5 asked Xint to minimize pre-distribution of details after a string of disclosure leaks, and Xint gave its own customers only a 24-hour notice that a critical patch was coming, without naming the target. The speed at which AI-enabled attackers can respond and reverse-engineer exploits from even the smallest clues means disclosure processes established over decades are changing, said Brian Pak, co-founder and CEO of Xint, who called nginx2root even more exploitable with broader impact than PHP2Shell, the company's earlier Linux stack finding.

For administrators the message is blunt: update to nginx 1.31.9 or 1.30.7 immediately, and remember that disabling HTTP/2 only blocks two of the five flaws. Distributions may backport fixes without changing the version number, so the full package version needs checking. The disclosure landed one day after Java4Shell, the OpenJDK flaw disclosed with the help of an AI agent, and it hardens what is becoming this year's most uncomfortable security trend: the machines finding the next nginx are already running.

Comments (0)

Log in to join the discussion

Log In

No comments yet