OpenAI has disclosed two critical memory-safety vulnerabilities in PHP, the scripting language that quietly runs a huge share of the web. Both flaws could let an attacker run code on a server without logging in first.
What OpenAI found
The two vulnerabilities are tracked as CVE-2026-109618 and CVE-2026-109619. Each carries a CVSS severity rating of 9.2, which places it firmly in the critical category.
Both bugs are memory-safety issues. In practice, a clever attacker can abuse that sloppiness to slip in instructions of their own. Here, the result is unauthenticated remote code execution. Put simply, a stranger on the internet could potentially take control of an affected server without needing a password.
Fixes were released on October 21, 2026. The patched versions are PHP 8.2.35, 8.3.36, 8.4.27, and 8.5.12.
Building those emergency patches took approximately two months. The work involved PHP maintainers collaborating with outside security firms, with PHP leading the disclosure process.
A checker tool, not an exploit
Alongside the disclosure came a tool called php2shell-checker. It lets site operators assess whether their setup is vulnerable without sending exploit traffic to the target.
OpenAI’s Codex, its AI coding agent, can also check whether a website’s PHP version is vulnerable to these recent flaws.
Why this one is different
The research was funded through OpenAI’s Patch the Planet initiative, which began in August 2026. The audit that surfaced these bugs was part of that program.
According to the research findings, this marks the first public instance of advanced AI identifying vulnerabilities at internet scale before they could be exploited.
The research findings state that 70% of websites use PHP. Even if only a slice of those run the affected versions, the potential blast radius is enormous.
What this means
For anyone running PHP in production, the action item is simple and urgent. Check the version, and if it predates the patched releases on its branch, upgrade.
The fixes cover the 8.2, 8.3, 8.4, and 8.5 lines. Anyone outside those lines has an uncomfortable conversation with their infrastructure team ahead of them.
The disclosure process involved security experts from PHP, Calif, and Google’s Wiz team, who utilized AI-driven tools to expedite the uncovering and hardening of these vulnerabilities before they were publicly announced.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.