The PHP project this week released emergency patches for two critical vulnerabilities, collectively nicknamed PHP2Shell, that AI models played a central role in uncovering. Security press coverage, including BleepingComputer, CSO Online and Fortune, reports that both are unauthenticated remote code execution flaws.
The vulnerabilities
The bugs are tracked as CVE-2026-109618 and CVE-2026-109619, each reported with a CVSS v4 score of 9.2. Fixes ship in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Reporting says the flaws affect PHP installations that serve web requests, including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed. The PHP security team reportedly rated mod_php as the highest risk, citing a demonstrated exploit in which an unauthenticated request can trigger the attack before any application code runs. The project called updating “the only safe choice”. PHP is widely cited as running on roughly 70% of websites.
The role of AI
According to press reports, security firm Calif found the first flaw during an audit of PHP’s source code funded through OpenAI’s Patch the Planet initiative, using OpenAI’s GPT-5.6 Sol model. Wiz says it independently found the flaws weeks later using its Wiz Atlas tool together with Google’s Gemini 4 Argon and GPT-5.6 Sol. Outlets differ on exactly when the first bug was found, so this article does not give a date.
OpenAI described the issue as “one of the most impactful security issues in recent memory” and warned that the same capabilities can help attackers develop exploits quickly. It is also sharing a PHP version checker as a Codex skill so organizations can identify vulnerable versions and apply the correct update.
What to do
Administrators running PHP on web-facing servers should move to one of the patched releases for their branch. The episode illustrates a point defenders have raised repeatedly: models that find bugs before attackers do can shorten the window for fixes, but the same tools lower the cost of building exploits once a patch reveals the flaw.