Running anything on PHP? An OpenAI model helped find two critical PHP flaws, so upgrade to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 today

#What changed

On Oct 21, OpenAI said its models helped uncover two critical flaws in PHP, the language behind a large share of the web. Per OpenAI's write-up, researchers at Calif found one of them using GPT-5.6 Sol as part of OpenAI's Patch the Planet program, and Wiz independently reported both. PHP's maintainers led the fix over roughly two months and released it the same day.

The PHP project's advisory says the two bugs (CVE-2026-109618 and CVE-2026-109619) can allow unauthenticated remote code execution. Fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35.

#Who is affected

The PHP advisory lists every PHP install that serves web requests: PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed. It rates mod_php as highest risk because an attack can fire before any of your application code runs. If you have used Codex or ChatGPT to build or deploy a WordPress, Laravel or other PHP site, this applies to you. PHP 8.1 and older no longer get official updates.

#What to do now

1. Check which PHP version each of your servers runs (php -v, or your host's dashboard).
2. Update to 8.5.12, 8.4.27, 8.3.36 or 8.2.35. On a Linux distro or managed host, take your vendor's patched package, since PHP says vendors got the fixes in advance.
3. On 8.1 or older, upgrade to a supported branch or ask your LTS vendor for a backport. PHP also posted minimal source backports for 7.4, 8.0 and 8.1.
4. If you cannot update today, PHP lists stopgaps such as a low post_max_size and request-size limits at the proxy. It says plainly that these do not make a server safe.
5. Asking Codex to do the check? Have it find every PHP version across your repos, Dockerfiles and deploy configs, then compare them with the list above. Review its changes before you deploy.

#What is not confirmed

OpenAI says it is sharing a PHP version checker and will launch a Codex skill to help find vulnerable versions. That skill was not out when we checked. OpenAI's claim that AI found the bugs before attackers did is its own statement. We have not tested the exploit, and we do not know of attacks in the wild yet.