AI helped find two critical holes in PHP, the language behind much of the web, and governments pulled sites offline before the fix landed

PHP's maintainers released emergency security updates on Wednesday, Oct. 21, 2026, for two critical flaws that can let an attacker run code on web servers without logging in, after OpenAI said researchers using its GPT-5.6 Sol model found one of them and Google's Wiz said it independently found both.

This is the AI-security future people have been warning about, arriving on a Wednesday. Frontier models helped find a decade-old flaw in software that runs a huge share of the web, which is the good news: defenders got there first. The bad news is everything after that. The details leaked, governments had to yank their own websites offline before a patch existed, and OpenAI says more bugs like this are already in the queue. AI makes finding holes cheap; fixing them still runs on unpaid open-source maintainers and IT teams applying updates one server at a time. Expect OpenAI's disclosure "framework" to get a lot of scrutiny from people who did the hard part.

PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35 fix two critical vulnerabilities that can allow unauthenticated remote code execution. That means an attacker on the internet can run their own code on the server without a password. CVE-2026-109618 is a case where running out of memory can leave a request's data only partly set up. CVE-2026-109619 is a case where old internal state survives after its memory is freed and a later request can reuse it. A CVE is the public ID for one specific flaw. PHP says every installation that serves web requests is at risk, including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP, and LiteSpeed. The risk is highest for mod_php, PHP built into the Apache web server, where a demonstrated exploit works before any of the site's own code runs. Both flaws are rated 9.2 on CVSS v4, The Next Web reports. CVSS is a severity score that tops out at 10, so 9.2 is near the top. Because of the severity, PHP also published minimal source backports for the unsupported 7.4, 8.0, and 8.1 lines. PHP says only updating is a complete fix. Those lines are PHP's, and the score is The Next Web's.

People at the security firm Calif, at Google's Wiz, and on PHP's maintainer team did the finding and the fixing. The models were tools in that work. OpenAI says Calif found a critical flaw using GPT-5.6 Sol as part of OpenAI's Patch the Planet initiative, and that Wiz independently discovered and reported both flaws. PHP's maintainers led the fixes. OpenAI, Wiz, Cloudflare, and others helped with research, testing, and coordination over the past two months. The Next Web says Wiz used its Wiz Atlas tool with Google's Gemini 4 Argon and with GPT-5.6 Sol. Calif said it used only generally available models, and that knowing PHP's history mattered more than which model was in the loop: "No model is simply capable of unearthing 'all bugs' with simple prompts. At least for now." Those lines are OpenAI's, The Next Web's, and Calif's.

OpenAI called this "one of the most impactful security issues in recent memory," and said the bugs are "among several significant vulnerabilities we've already found and are working to fix." It said similar findings should "increase sharply in the coming months." It proposed a five-stage disclosure process: validate and fix; give affected organizations time; coordinate with governments; prepare updates across products; publish guidance. It said it worked with the White House's Super Intelligence Force, the Office of the National Cyber Director, and CISA, the U.S. cyber agency. It is sharing a PHP version checker as a Codex skill, a packaged check its coding agent can run. This disclosure sits apart from the $1 billion Daybreak pledge OpenAI announced earlier for frontline defenders. Those lines are OpenAI's.

The fix moved up because the details leaked. The Next Web, citing the Wall Street Journal, reported that details got out before the scheduled release and a proof-of-concept circulated, so PHP brought the update forward. A proof-of-concept is a working demonstration of the attack. CERT-EU, the European Union's computer-emergency team, and Britain's National Cyber Security Centre advised taking systems offline where they could not be patched at once. The Associated Press reported that the U.S. National Security Agency and the FBI issued warnings. Finland temporarily shut down public government web services. The Dutch government took sites including Overheid.nl, MijnOverheid.nl, and DigiD offline as a precaution, its own notice says. No public count of confirmed mass exploitation was published.

In plain terms, PHP on Wednesday shipped emergency updates for two critical flaws that can let an outsider run code on a web server without logging in. Calif found one with OpenAI's GPT-5.6 Sol. Wiz says it found both on its own, using AI tools. PHP's maintainers wrote the fixes, and PHP says installing the update is the only complete fix. Governments, including the Netherlands, took public sites offline while they patched. OpenAI says more bugs of this kind are already in the queue.