What happened
On Oct 21, the PHP project released PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to fix two critical bugs, CVE-2026-109618 and CVE-2026-109619. The PHP project says they can let an attacker run code on a server without logging in, and that every PHP installation serving web requests is at risk. The researchers' site, php2shell.com, says PHP powers about 70% of websites, including WordPress, Drupal and MediaWiki, and that a site can be affected even when its own code is secure.
The bugs were found with AI help. OpenAI says the security firm Calif found a critical flaw using GPT-5.6 Sol, and that Wiz found and reported both on its own. The researchers' site says Wiz also used Google's Gemini 4 Argon, and that the PHP team worked with the researchers on the fixes.
Why updating WordPress isn't enough
The bug is in PHP itself, under your app. php2shell.com says the only complete fix is installing a fixed PHP release on every runtime, then restarting Apache, PHP-FPM and other long-running PHP workers. It also says updating an application doesn't update PHP. On shared or managed hosting, that work belongs to your host.
The PHP project lists stopgap settings for servers that can't update yet, but it says they don't make an installation safe. On Oct 23, it added one more: turn off session upload progress until the next bugfix release, against an attack path it calls unlikely.
How to check your site
OpenAI published a free checker, usable as a Codex skill or a Node script. It reads the PHP version a site advertises and reports OK, POTENTIALLY VULNERABLE or UNKNOWN, without sending attack traffic. A version number isn't proof, though. Many sites hide it, and Linux vendors ship fixes under older-looking numbers. The reliable step is asking your host to confirm the fixed version and the worker restart in writing. Our new playbook walks through that, with prompts to draft the message.
What's unconfirmed
None of the official pages we read say whether these bugs are being exploited in the wild. We also can't tell you whether a given host has patched. Only the host can confirm that. OpenAI expects findings like this to increase sharply in the coming months, so it's worth having a routine ready for the next alert.