On October 21, the NSA's cybersecurity directorate advised all organizations using PHP to prepare "effective immediately" for an update addressing two critical vulnerabilities, and OpenAI said its models helped researchers find them. PHP maintainers have since released the fixes. OpenAI calls the bug "one of the most impactful vulnerabilities in recent memory" and says it marks "a new phase, where AI can find security bugs with internet-scale impact."
The NSA's warning
The NSA said PHP is widely used for web applications and that these vulnerabilities often provide a direct path to the server's file system or database. If an immediate update is not feasible, it recommends taking the system offline until it can be fully mitigated, and it promised more specific guidance shortly.
How OpenAI says it was found
OpenAI says its models helped researchers, working as part of its Patch the Planet initiative, discover critical flaws in PHP, the open-source language it says runs an estimated 70% of websites. Over the past two months it led a coalition of maintainers, researchers and security providers to prepare fixes and protections before disclosure. The duration was necessary because the fix was intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade, in addition to applying mitigations from security vendors. OpenAI's write-up proposes a staged framework for coordinated disclosure of critical flaws in widely used software.
Coverage and vendor posts fill in the technical picture. The flaws, named PHP2Shell, are tracked as CVE-2026-109618 and CVE-2026-109619, rated 9.2 on CVSS v4.0, and allow unauthenticated remote code execution. Wiz, whose researchers found and reported both, says every PHP release of the past decade is affected by at least one, and demonstrated code execution against the latest WordPress, Drupal and Joomla container images. Researchers at Calif and OpenAI also reported the flaws. Fixes shipped in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35, with source backports for end-of-life 7.4 through 8.1, according to Help Net Security, while 5.4 through 7.3 get no official fix. Greg Brockman said details leaked early, so the release moved up from October 22 to October 21, and that the bug had been in PHP for 14 years. Reports say exploitation attempts have begun.
The Defender's Window
OpenAI frames this as "the Defender's Window in action": to its knowledge, frontier AI found the vulnerability ahead of attackers, giving defenders time to prepare. It is the case behind the argument in Why Every Lab Suddenly Has a 'Cyber' Model and the open letter calling for a global cyber-defense surge, and it gives weight to OpenAI's decision to pause its biggest training run over cyber capability. OpenAI shares a PHP version checker as a Codex skill and reportedly commits $1 billion of subsidized Daybreak access for organizations with limited security resources, which is not a cash grant. The window was short, though: the leak forced the release a day early.