AI Finds Decades-Old Bugs in PHP, NSA Says Pull Sites If Unpatched

By Grace Kim

Two critical vulnerabilities that have sat in PHP since version 5 shipped in 2004 now threaten roughly 70% of websites on the internet — and the U.S. National Security Agency said Wednesday that any site that can't patch immediately should be pulled off the web until it can. The PHP Foundation has already issued patches.

By some counts, the affected footprint exceeds 33 million sites. PHP, a server-side programming language, powers the bulk of the web's applications and underpins popular content management systems such as WordPress.

The two holes allow attackers to infiltrate a server's file system or a database used by a web application. The NSA's warning was blunt about severity: patch now or go dark.

How old is the exposure?

The flaws have been tucked away in every version of PHP since version 5, released in 2004. Sites not running relatively current versions face extra work, because PHP 8.1 and older versions no longer receive support. The recommendation: update web servers to the current release, PHP 8.5.12.

In plain terms — if your server is running stale PHP, the fix isn't optional and it isn't partial.

Who found the bugs?

AI did. Security testing company Calif, engaged in OpenAI's "Patch the Planet" initiative to find flaws in open-source software, discovered one of the two flaws using GPT-5.6 Sol, OpenAI disclosed in a blog post. Separately, Google's Wiz team deployed its own AI stack — the Wiz Atlas security AI agent, Gemini 4 Argon, and GPT-5.6 Sol — and found multiple vulnerabilities in PHP.

Both OpenAI and Wiz followed responsible disclosure practices. They worked with PHP on patches before going public, then collaborated to flesh out findings and test fixes, drawing in a host of outside helpers. Wiz publicly thanked Automattic, Akamai, AWS, Cloudflare, Fastly, Imperva, Microsoft, and Vercel.

WordPress, for its part, issued a statement saying the flaw isn't WordPress-specific. Sites running its CMS must patch their servers and work with their web-hosting companies to ensure protection.

Exploited in the wild?

Not yet. OpenAI says researchers found the flaws before hackers did, and there were no reports of exploits in the wild. Part of the reason: before AI, the bugs were simply too hard for attackers to weaponize.

"With AI this is no longer a barrier. LLMs can help adapt an exploit to different PHP versions, configurations and memory layouts," Wiz warned in its blog post on the flaws.

That quote cuts both ways. This is the kind of security issue discovered and prevented because of AI — and a high risk only because of AI.

What happens next?

The disclosure puts hackers fully in the know. The race is now on for the internet to patch tens of millions of websites before attackers hit the unpatched ones. That urgency is exactly what drove the NSA's guidance: if you can't patch your PHP site right away, you're better off taking it offline until you can.