On October 21, 2026, the U.S. National Security Agency (NSA) warned of a serious threat to internet infrastructure: two critical vulnerabilities had been found in the PHP programming language. The PHP Foundation has already issued fixing patches. The flaws had been hiding in every version of the language since PHP 5, released in 2004 β meaning nobody noticed them for over twenty years. Most remarkably, they were found not by people, but by artificial intelligence systems.
NSA: if you can't patch, temporarily take the site offline
The NSA rated the vulnerabilities so severely that it advised websites unable to apply a patch immediately to temporarily disconnect from the internet until they can. Such a harsh recommendation at the level of a government agency is rare β it shows how serious the flaws are.
The holes let attackers access the server's file system or the database used by the web application. In practice that means full site takeover: files can be altered, data stolen, or the server itself turned into a springboard for further attacks. PHP powers roughly 70% of internet websites β over 33 million sites; popular content management systems like WordPress are also built on PHP. Ancient defects in such a widespread language leave millions of sites vulnerable at once.
PHP 8.1 and older versions are no longer supported, so sites running them are left without protection unless they update. The current release is PHP 8.5.12. The team maintaining PHP asked web developers to prepare for two emergency security patches.
Who found the flaws and how
One flaw was discovered through OpenAI's "Patch the Planet" initiative: security testing firm Calif found it using the GPT-5.6 Sol model, and OpenAI disclosed the work in its blog. The initiative's goal is to find vulnerabilities in the open-source ecosystem with AI and disclose them responsibly.
Wiz independently found several PHP vulnerabilities using its Wiz Atlas security AI agent and the Gemini 4 Argon and GPT-5.6 Sol models. Both teams followed responsible research practice: before publicly disclosing the flaws, they worked with the PHP team on patches and then jointly verified the findings. This is an important detail β the researchers did not disclose the flaws directly, but waited for patches to be ready.
Among the companies Wiz thanked are Automattic, Akamai, AWS, Cloudflare, Fastly, Imperva, Microsoft and Vercel β these companies were notified before the patch rollout and prepared their infrastructure. WordPress said in its statement that the flaw is not WordPress-specific, but sites running its CMS should update servers and secure protection together with hosting providers.
"With AI this is no longer a barrier. LLMs can help adapt an exploit to different PHP versions, configurations and memory layouts" (Wiz, vulnerabilities blog post).
Why now, and why the risk is high
According to OpenAI, the flaws were found by researchers before hackers, so no exploits in the wild have been reported yet. Before AI, exploiting these flaws was very hard for hackers: finding complex defects in a twenty-year-old codebase by hand was nearly impossible.
But Wiz warns that AI makes these vulnerabilities even more dangerous: large language models can adapt attack code to different PHP versions, configurations and memory layouts. Previously an attacker had to write a separate exploit for each version β that barrier is now gone. In other words, this problem was found and prevented thanks to AI β but it creates high risk precisely because of AI.
Once the issue was publicly disclosed, hackers learned of it too. Now the race is on: the security community recommends updating sites before unpatched ones become attack targets. History shows that with such widespread flaws, exploitation attempts begin within days of a patch release.
Patches are out β next step
The PHP Foundation issued fixing patches, and the PHP maintenance team asked developers to prepare for two emergency patches. The recommendation is clear: update the web server to the current PHP version (8.5.12). PHP 8.1 and older are no longer supported β sites on such versions have no protection without an update. The NSA's harshest recommendation is for those delaying patching: temporarily take the site offline until it can be fixed. Detailed technical analysis is in the TechCrunch and Ars Technica write-ups.




