On October 21, 2026, the PHP project released PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to fix two critical vulnerabilities that can allow unauthenticated remote code execution, and urged all users to update immediately. CVE-2026-109618 is RCE via out-of-memory corruption: out-of-memory handling can interrupt PHP while internal request data is only partly initialized, and cleanup then treats it as valid. CVE-2026-109619 is RCE via stale global state: request-owned state can outlive its freed memory and be used by a later request on the same worker. A pre-announcement on October 20 rated both CVSS v4.0 9.2. PHP says all installations serving web requests are at risk - PHP-FPM, Apache mod_php, php-cgi, FrankenPHP and LiteSpeed - with risk highest for mod_php, where a demonstrated exploit fires before any application code runs. Because of the severity, the security team also published minimal source backports for the unsupported PHP 7.4, 8.0 and 8.1 branches.
The same day, OpenAI published “Coordinated disclosure of critical vulnerabilities,” saying it had spent two months leading a coalition of maintainers, researchers and security providers on the PHP response. Under its Patch the Planet initiative, Calif identified a critical flaw using GPT-5.6 Sol, and Wiz independently discovered and reported both vulnerabilities; PHP’s maintainers led the fixes. OpenAI called it “one of the most impactful security issues in recent memory,” said frontier AI found the bugs before attackers did, and said these were among several significant vulnerabilities it is working to fix, with similar findings expected to “increase sharply in the coming months.” It described engagement with the White House’s Super Intelligence Force, ONCD and CISA, and proposed a five-stage framework for coordinating fixes to critical bugs in widely used software. On October 23 PHP added an informational advisory for a related, less likely RCE vector, CVE-2026-115560.
Why it matters: given what OpenAI calls PHP’s widespread use, this is an internet-scale, pre-authentication bug class in a core runtime where an AI model is credited in the discovery. It turns the “AI finds bugs” story from counts in vendor reports into an emergency patch cycle that operators everywhere had to act on.
What it does not show: PHP’s own advisory credits no reporter, so the AI attribution rests on OpenAI’s account, and Wiz found both bugs independently by means not described. Neither source reports exploitation in the wild, and how much of the discovery was model output versus human analysis is not broken out.