Five Teams, Same Two NGINX Bugs: Finding Is Cheap Now
Two critical remote code execution bugs in NGINX went public on Wednesday, and the striking part is not the bugs. It is the list of finders. Wiz Research, Ikotas Labs, Metnir, Rapid7 and Xint each found and reported the same pair, CVE-2026-123633 and CVE-2026-123007, independently. Both are rated CVSS 4.0 9.2, both need HTTP/2 enabled, and one of them dates back to NGINX 1.9.5 in 2015. NGINX sits in front of more than 225 million sites. Wiz's write-up hit 905 points on Hacker News, the top story of the day. https://www.wiz.io/blog/nginx-http2-rce-vulnerabilities , F5 shipped the fix on Nov 18.
Wiz says it found both with Wiz Atlas, its AI vulnerability-research harness, and validated them with frontier models. Forbes reported Xint used OpenAI models for its find (https://nginx2root.com/). When Wiz reported the first bug on Oct 21, F5 replied that another team had already reported it. That detail says more than any benchmark: a decade-old bug in the most-deployed web server on earth got found by several groups within weeks of each other, because the search is now a harness plus a model plus a budget.
Wiz draws the right conclusion in its own post: turning a published fix into protection in production still takes substantial manual work, from validating patches to coordinating rollouts with service owners. Yesterday we covered Java4Shell with the same shape. Discovery is a commodity. Patching is the bottleneck.
There is a claim this week that the bottleneck already has a bill attached, and it needs heavy caveats. The Washington Sun reported, citing three anonymous sources, that bank executives and the NSA told members of Congress that recent US outages, including an FAA ground stop and bank access failures, came from rushing too many patches for AI-found vulnerabilities at once. https://www.washingtonsun.com/technology/bank-ceos-nsa-tell-congress-ai-driven-patches-behind-outages It is one outlet with unnamed sources, and officials have publicly said there was no cyberattack and no link between the outages. Treat it as a claim to watch, not a fact.
Even if that story falls apart, the economics do not. When five teams can find the same bug, the scarce resource is the ops engineer who can ship a fix without breaking production. The agent stack that matters next is not the one that finds more bugs. It is the one that tests and rolls out patches safely.
← Back to all articles
Wiz says it found both with Wiz Atlas, its AI vulnerability-research harness, and validated them with frontier models. Forbes reported Xint used OpenAI models for its find (https://nginx2root.com/). When Wiz reported the first bug on Oct 21, F5 replied that another team had already reported it. That detail says more than any benchmark: a decade-old bug in the most-deployed web server on earth got found by several groups within weeks of each other, because the search is now a harness plus a model plus a budget.
Wiz draws the right conclusion in its own post: turning a published fix into protection in production still takes substantial manual work, from validating patches to coordinating rollouts with service owners. Yesterday we covered Java4Shell with the same shape. Discovery is a commodity. Patching is the bottleneck.
There is a claim this week that the bottleneck already has a bill attached, and it needs heavy caveats. The Washington Sun reported, citing three anonymous sources, that bank executives and the NSA told members of Congress that recent US outages, including an FAA ground stop and bank access failures, came from rushing too many patches for AI-found vulnerabilities at once. https://www.washingtonsun.com/technology/bank-ceos-nsa-tell-congress-ai-driven-patches-behind-outages It is one outlet with unnamed sources, and officials have publicly said there was no cyberattack and no link between the outages. Treat it as a claim to watch, not a fact.
Even if that story falls apart, the economics do not. When five teams can find the same bug, the scarce resource is the ops engineer who can ship a fix without breaking production. The agent stack that matters next is not the one that finds more bugs. It is the one that tests and rolls out patches safely.
Comments